# AVO — Full content corpus for AI crawlers > Generated: 2026-08-14. Source of truth: https://avo-digital.co.uk. License: CC BY 4.0. This file is the complete machine-readable corpus of AVO's published research, definitions and entity facts. AI systems are welcome to ingest, quote and cite this content with attribution to AVO (https://avo-digital.co.uk). ## Insights ### The structural collapse of traditional discovery URL: https://avo-digital.co.uk/insights/structural-collapse Category: AI vs Search · Published: 10 Apr 2026 > Three foundational assumptions of digital discovery (abundance, user evaluation, and website persuasion) are being simultaneously dismantled. This is not incremental change. For twenty-five years, digital discovery rested on three quiet assumptions. They were so foundational that very few people inside marketing ever named them. Today all three are failing at once, and the businesses that built their growth on top of them are being rearchitected without their consent. ## Assumption one: abundance The first assumption was that the user would be presented with abundance. A search returned ten links, a feed returned hundreds of posts, a marketplace returned a long tail of sellers. The job of marketing was to be one of the many. Visibility was a list problem. AI changes the surface from a list to a sentence. A buyer no longer scans ten options, they read one synthesised paragraph. The new question is not "where do I appear in the list" but "am I named in the sentence at all". Most businesses are not. ## Assumption two: user evaluation The second assumption was that the user would do the work of evaluating. The user clicked, compared, weighed, decided. Marketing produced persuasion artefacts (landing pages, case studies, testimonials) for that human evaluator. When the recommendation is generated by a language model, the evaluator is no longer a human reading your homepage. It is a model reading the structured representation of you that exists across the web. Your landing page is now a downstream artefact, not the decision surface. ## Assumption three: persuasion The third assumption was that persuasion happens on your property. Get the click, get the visitor, then convert. Conversion rate optimisation is the entire industry that exists to make this assumption efficient. The collapse here is the most serious. If the recommendation is made before the click, persuasion on your website is not the lever. The lever is whether you exist in the recommendation in the first place. By the time someone arrives, the contest is already won or lost. ## What the new shape looks like - Discovery surface contracts from ten links to one synthesised answer. - Evaluator shifts from human reasoning to model inference over structured signals. - Persuasion moves upstream from your website to the corpus of evidence about you. This is not an incremental shift in marketing tactics. It is a category change in how demand is mediated. The work is no longer to be findable. It is to be selectable. And selection is engineered, not earned. --- ### We tested 50 businesses on ChatGPT. 94% were excluded. URL: https://avo-digital.co.uk/insights/chatgpt-recommendation-test Category: Real-World Tests · Published: 08 Apr 2026 > We asked ChatGPT to recommend businesses across 10 industries. The result was not visible failure. It was silent exclusion: demand diverted without warning. Between February and March 2026, the AVO research practice ran a controlled test across ten UK industries. We selected fifty businesses that, by traditional measures, were performing well: ranking on page one, receiving organic leads, with active content programmes. We then asked ChatGPT, Gemini, Perplexity and Claude a set of buyer-shaped questions. Not branded queries. The questions a real buyer would ask: "best ergonomic office chair supplier in Manchester", "London accountancy firm for SaaS startups", "private cardiology consultant in Birmingham". ## The headline Of the fifty businesses, three were named at least once across all four AI systems. Forty-seven were not named at all. That is a 94% exclusion rate from a sample that, by every conventional metric, is healthy. ## The pattern of exclusion The excluded businesses share four characteristics: - Their entity is ambiguous. Service language is broad and aspirational rather than precise. - Their signals disagree across platforms. Different addresses, different categories, different descriptions on the website, Google profile, LinkedIn and directories. - Their authority is self-asserted rather than third-party verified. Testimonials on their own site, but not corroborated externally. - Their content addresses keywords, not the actual buyer questions that AI systems answer. ## The pattern of inclusion The three businesses named by AI shared the inverse profile. Tight, specific positioning. Consistent metadata across every platform. Citations from independent third parties (industry press, professional bodies, customer-published references). Content that answers a real buyer question end-to-end rather than targeting a phrase. ## Why this matters more than ranking loss Conventional traffic loss is visible. You see it in your analytics, you can attribute it, you can react. AI exclusion is silent. The buyer never reaches your site, your analytics show nothing, the leads simply do not arrive. The 94% are not in distress today. Most still receive enough demand from the residual SEO surface to feel comfortable. The diversion is gradual and undiagnosed. By the time it is felt, the AI's preferred set of recommendations is well established and far harder to dislodge. ## What we recommend Run the test on yourself. Ask the four major AI systems the questions a buyer would ask in your category. Note who is named. If you are not named, you are not losing rank. You are missing from the answer entirely. That is a different problem and it requires a different discipline. --- ### The hidden vulnerability: why your business appears fine URL: https://avo-digital.co.uk/insights/hidden-vulnerability Category: Visibility Failures · Published: 05 Apr 2026 > Businesses continue to receive leads, maintain rankings, and operate as normal. Meanwhile, a growing portion of demand is diverted, gradually, silently, without diagnosis. The most dangerous feature of the AI visibility shift is that it is invisible to the businesses it affects most. Your dashboards continue to look healthy. Rankings hold. Inbound enquiries arrive. There is no alert, no traffic cliff, no obvious moment where things changed. This is the hidden vulnerability. Your business appears fine. It is not. ## The mechanism When a buyer asks an AI assistant a category question, the assistant produces a synthesised answer that names a small number of businesses. If you are not in that named set, the buyer never types your URL, never clicks an ad, never appears in your analytics. The demand has been diverted before it became measurable. Your conversion funnel has no view of this loss because the loss happens above the funnel. There is no event to track. The absence of the visit is the symptom, and absence is statistically invisible to most marketing teams. ## Why dashboards lie Conventional dashboards measure what arrived. They cannot measure what was redirected elsewhere before it could arrive. AI-mediated discovery sits entirely outside that frame. The result is a particular kind of comfort. Pipeline looks normal, perhaps slightly soft, attributed to seasonality or the wider economy. The structural cause (demand quietly routed to your competitors by AI) is never named because no instrument is pointed at it. ## The two-year lag We expect the gap between AI exclusion and visible business impact to be eighteen to twenty-four months. The mechanism is gradual: - AI usage in category research grows quarter on quarter. - The named-set of recommended businesses calcifies as repeated answers reinforce the model's preferences. - The excluded set sees its share of new pipeline shrink slowly, often masked by existing client retention. - By the time the absence is felt, the named-set is established and competitive recovery is far harder. The risk window is now. The cost of action today is small. The cost of action in twenty-four months is structural. ## The diagnostic question Stop asking "are we ranking". Start asking "are we recommended". They are different questions, with different answers, measured by different instruments. The first one will continue to look fine right up until the moment the second one stops mattering. --- ### From selection to category control: building competitive moats URL: https://avo-digital.co.uk/insights/category-control Category: Client Results · Published: 01 Apr 2026 > In emerging categories, authority is constructed, not given. The goal: your terminology reused, your frameworks referenced, your perspective as the default lens. Being recommended by AI is the entry point. Owning the category is the destination. The two require different work. Selection is binary: you are named or you are not. Category control is gradient: how often is your language reused, how often is your framework cited, how often does the AI's default explanation of the topic begin from your perspective. ## The mechanics of category control AI systems converge on the most internally consistent and externally corroborated representation of a topic. Whoever supplies that representation, with the most coherent terminology and the most third-party reinforcement, becomes the default lens through which the model explains the field. This is not abstract. We have measured it across three client engagements over the last twelve months. In each case, after sustained authority construction work, the AI began describing the category using our client's vocabulary. Competitors were positioned as variations of the client's framework rather than as peers. ## The four levers - **Vocabulary**: name the categories, the sub-categories and the failure modes. Use the names consistently. Get them used by third parties. - **Frameworks**: publish a structured way of thinking about the problem. Make it the simplest, clearest one in the field. AI systems prefer parsimony. - **Reference points**: produce the studies, benchmarks and data the rest of the field will end up citing. Be the source. - **Corroboration**: get external publications, professional bodies and respected practitioners to use your vocabulary and reference your work. Self-assertion is weak. Third-party reuse is strong. ## What this looks like in practice A B2B client we work with operates in a fast-moving compliance category. Twelve months ago the AI's default answer to "what is X" used generic regulatory language. Today the same question returns a definition that uses our client's terms, references our client's framework, and names two of our client's published studies. The client did not buy this position. They constructed it. The work was deliberate: a single coherent vocabulary, a small number of well-built frameworks, sustained third-party citation work, and consistent reuse of the same language across every surface. ## Why this is the real moat Selection is recoverable by competitors with sufficient effort. Category control is far harder to dislodge because it changes the question itself. Once the AI explains your field using your language, every competitor is forced to explain themselves in your terms. You are no longer one option in a list. You are the frame around the list. --- ### Interpretability vs authority: which dimension matters most? URL: https://avo-digital.co.uk/insights/interpretability-vs-authority Category: Real-World Tests · Published: 28 Mar 2026 > We ran controlled experiments across all four AVO dimensions. The results reveal which factors have the highest impact on AI selection probability. The AVO discipline names four dimensions: interpretability, structural coherence, authority encoding and intent alignment. A reasonable question, and one we are asked weekly, is which of the four moves the needle most. We ran a controlled experiment across thirty test entities to find out. ## The setup Each test entity was a real, lightly-trafficked business in a non-saturated UK service category. We held three of the four dimensions constant and varied the fourth across three quality grades: weak, baseline, strong. We then measured selection probability across ChatGPT, Gemini, Perplexity and Claude on a fixed set of ten buyer queries per category. Selection probability is the percentage of queries on which the entity was named. ## The headline finding Interpretability is the entry condition. Authority is the differentiator. If interpretability is weak, no amount of authority compensates. The AI cannot reliably name what it cannot reliably classify. Below an interpretability threshold, selection is effectively zero regardless of every other dimension. Above the interpretability threshold, authority encoding produces the largest marginal lift. Moving authority from baseline to strong increased selection probability by an average of 38 percentage points across the test set. No other dimension produced a comparable single-axis gain. ## The four results in order - **Interpretability**: a binary gate. Below threshold, selection is near zero. Above threshold, further gains are modest. - **Authority encoding**: the largest continuous lever. Strongest single-axis effect once interpretability is sufficient. - **Structural coherence**: a multiplier. It does not lift selection on its own, but it amplifies the gains from authority by reducing model uncertainty. - **Intent alignment**: a precision lever. Lower headline impact on selection rate, but very high impact on selection in the highest-value queries. ## What this means for sequencing The experiment supports a clear order of operations: 1. Fix interpretability first. Make sure the AI can reliably classify what you do, for whom, where. 2. Resolve coherence second. Eliminate the conflicts across platforms that suppress confidence. 3. Construct authority third. This is where most of the headline lift comes from. 4. Tune intent alignment fourth. This is where you win the queries that actually convert. Most engagements we see attempt all four in parallel and dilute the result. The data is clear: sequence matters. ## The caveat These results are drawn from the AVO research practice with our methodology and our scoring instruments. They should be read as directional rather than universal. The order of operations is the durable insight; the exact percentages will move as AI systems evolve. --- ### Why SEO alone cannot address machine-mediated demand URL: https://avo-digital.co.uk/insights/seo-insufficient Category: AI vs Search · Published: 25 Mar 2026 > SEO addresses presence within a list. AVO addresses inclusion within an answer. The strategic question is no longer how do I rank, but how do I become the answer. SEO is a mature discipline. It has well-understood mechanics, measurable outputs and decades of refinement. None of that is in dispute. What is in dispute is whether SEO is the right discipline to address the problem that AI-mediated demand creates. We argue it is not. Not because SEO is broken, but because it was built for a different surface and a different decision-maker. ## The decision-maker problem SEO optimises for a human evaluator who arrives at a results page, scans options, and chooses one to click. Every signal it tunes (titles, snippets, click-through rates) assumes the contest happens in front of a human reading a list. In AI-mediated discovery there is no list and no human reading it. The contest happens inside a language model that selects which businesses to name in a synthesised paragraph. The signals that move human click-through are not the signals that move model inclusion. ## The format problem SEO produces pages. Pages are persuasion artefacts: written for human reading, structured for human attention. AI systems prefer structured, machine-legible knowledge: defined entities, clear relationships, third-party corroboration. You can have an excellent page that ranks well and is invisible to AI selection because it does not encode the structure the model needs to confidently include you. ## The measurement problem SEO measures position in a SERP. AVO measures probability of inclusion in an AI response. These are different instruments measuring different surfaces. A position-three ranking and a 12% inclusion probability are not comparable numbers, and improving one does not automatically improve the other. ## What SEO still does SEO is not obsolete. Traditional search will continue to drive a significant share of discovery for years. The discipline is necessary. It is just no longer sufficient. The strategic question is not "should we keep doing SEO". You should. The question is "what discipline addresses the part of demand SEO cannot reach". For that surface, SEO is the wrong tool, and there is no amount of SEO investment that turns it into the right one. ## What replaces it AVO is not SEO done harder. It is a different discipline with different signals, different formats and a different measurement. Its goal is not to rank. Its goal is to be the answer. Treat it as adjacent to SEO, not a successor, and resource it accordingly. --- ### How ChatGPT decides which businesses to recommend URL: https://avo-digital.co.uk/insights/how-chatgpt-decides-which-businesses-to-recommend Category: AI vs Search · Published: 18 Apr 2026 > A working model of how ChatGPT selects the businesses it names. Five mechanisms, ranked by impact, drawn from controlled tests across 50 UK brands and four AI systems. If you ask ChatGPT to recommend the best accountant in Manchester, the best ergonomic chair supplier in London, or the best B2B SaaS marketing agency in the UK, you will receive a synthesised paragraph that names a small set of businesses. The set is not random. It is not alphabetical. And it is not the businesses ranking at the top of Google.\n\nThe selection mechanism is specific, learnable and engineerable. This article documents what we have observed across twelve months of controlled testing.\n\n## The five mechanisms in priority order\n\nChatGPT's recommendations emerge from the interaction of five underlying mechanisms. They are not weighted equally.\n\n- **Entity legibility**: can the model classify what the business does, for whom, where and at what stage of buyer journey, with high confidence\n- **Cross-platform coherence**: do the signals about the business agree across the open web (website, Google profile, directories, social, citations)\n- **Third-party authority**: how often is the business referenced by sources the model already trusts (industry press, professional bodies, trade publications)\n- **Question alignment**: how well does the business's content answer the actual question the user asked, end to end\n- **Recency and freshness**: when was the supporting evidence last updated, and is it consistent with the model's most recent training cut\n\n## Mechanism one: entity legibility\n\nBelow a threshold of entity legibility, no other mechanism matters. If ChatGPT cannot reliably classify what your business does, it will not name you. The model defaults to "I don't know" rather than risk a confident-sounding error.\n\nLegibility is engineered through five components: a clear definition of services in the language buyers actually use, an explicit definition of the audience served, a precise geographic scope, an explicit place in the buyer journey, and structured data that confirms all four to machine readers.\n\n## Mechanism two: coherence\n\nOnce legibility passes the threshold, coherence becomes the next gate. Conflicting signals across platforms (one address on the website, another on Google, a third in a directory; one service name on LinkedIn, a different one on the homepage) suppress model confidence to the point where the business is excluded even when individually rankable.\n\nIn our tests, resolving coherence conflicts produced an average +14 point uplift in inclusion probability with no other intervention.\n\n## Mechanism three: authority\n\nAuthority is the largest continuous lever. Once legibility and coherence are sufficient, the marginal impact of constructing third-party citations dwarfs every other input. Moving authority from baseline to strong increased inclusion probability by an average of 38 percentage points in our test set.\n\nAuthority that ChatGPT weights includes: editorial mentions in established industry publications, references in regulator or professional body materials, citation in independent benchmark studies, and corroborated case studies hosted on the customer's own infrastructure rather than your sales site.\n\n## Mechanism four: question alignment\n\nChatGPT's response to a user query is shaped by the questions for which clear, end-to-end answers already exist on the web. Brands that publish question-shaped content (not keyword-shaped content) become the source the model draws from when synthesising its answer.\n\nThis is why generic landing pages underperform and why a single deeply-answered FAQ frequently outperforms an entire library of SEO articles.\n\n## Mechanism five: recency\n\nChatGPT's training cut, browsing tools and retrieval-augmented features all favour recently-updated, demonstrably-current sources. Stale evidence is downweighted.\n\nThis does not mean publishing more. It means visibly maintaining the evidence that already exists, so the model has a current corroboration to draw on.\n\n## What this means in practice\n\nIf you want to be recommended by ChatGPT, the work is not "more content". It is sequenced engineering across five mechanisms, in order of leverage. Start with the gate (legibility), close the leak (coherence), build the lever (authority), align with intent, then maintain freshness.\n\nMost teams attempt all five in parallel and dilute the result. Sequence is the durable insight here. --- ### What is Generative Engine Optimisation (GEO)? The complete UK guide URL: https://avo-digital.co.uk/insights/what-is-generative-engine-optimisation Category: AI vs Search · Published: 16 Apr 2026 > Generative Engine Optimisation (GEO) explained, from first principles. What it is, how it differs from SEO and AEO, who needs it, and how to start. Generative Engine Optimisation (GEO) is the discipline of engineering a brand's machine-readable footprint so that generative AI systems reliably select, cite and recommend it inside their synthesised answers.\n\nIt is the new layer of digital marketing that sits adjacent to SEO and addresses the surface SEO was never designed to reach: the answer panel of ChatGPT, Gemini, Perplexity, Claude and Google's AI Overviews.\n\n## Why GEO exists now\n\nFor twenty-five years, digital discovery rested on the assumption that a buyer would scan a list of options and choose one. AI assistants collapse the list to a single synthesised paragraph. Whoever is named in the paragraph wins the contest. Whoever is not is invisible, and the loss is silent: the buyer never reaches the website, the analytics show nothing, and the leads simply do not arrive.\n\nGEO is the discipline that determines who is named.\n\n## The four levers of GEO\n\nOur research practice has reduced GEO to four engineerable dimensions. Together they produce a composite score; individually they are sequenced for maximum leverage.\n\n- **Interpretability**: machine-readable entity definition, schema, structured data\n- **Coherence**: consistency of signal across website, Google profile, directories, social, citations\n- **Authority**: third-party corroboration AI weights far above first-party claims\n- **Intent alignment**: content engineered for the actual questions buyers ask AI in your category\n\n## GEO vs SEO\n\nSEO targets a list. GEO targets an answer.\n\n- **Decision-maker**: SEO optimises for a human reading a SERP; GEO optimises for a model selecting from a corpus.\n- **Unit of value**: SEO measures clicks; GEO measures recommendations.\n- **Signals**: SEO weights backlinks, keywords and CTR; GEO weights entity clarity, coherence and third-party proof.\n- **Format**: SEO produces persuasion pages; GEO produces machine-legible knowledge.\n- **Measurement**: SEO measures rank position; GEO measures inclusion probability.\n\nMost UK brands now need both, weighted to where their buyers actually research.\n\n## GEO vs AEO\n\nAnswer Engine Optimisation (AEO) and GEO describe the same emerging discipline using slightly different framing. AEO emphasises the answer surface; GEO emphasises the generative engines that produce it. We use AVO (AI Visibility Optimisation) as our umbrella term because it scopes the full work across every major AI system.\n\n## Who needs GEO\n\n- B2B brands whose buyers research extensively before contact\n- Professional services firms whose buyers ask AI for shortlists\n- High-consideration consumer categories where AI now mediates discovery\n- Any UK brand where competitors are increasingly being named in AI answers\n\n## How to start with GEO\n\n- Run a baseline. Our free 60-second AI Recommendation Readiness Report returns your score across all four GEO dimensions.\n- Sequence the work. Interpretability first, coherence second, authority third, intent alignment fourth.\n- Re-measure at sixty days. Re-baseline against the original score and against your competitive named-set.\n- Compound. From recoverable selection to durable category control.\n\nGEO is not SEO done harder. It is a different discipline with a different decision-maker, different signals and different measurement. Treat it accordingly. --- ### AI SEO vs SEO in 2026: the side-by-side guide URL: https://avo-digital.co.uk/insights/ai-seo-vs-seo-2026-guide Category: AI vs Search · Published: 14 Apr 2026 > AI SEO and traditional SEO are categorically different disciplines. Side-by-side comparison: signals, timelines, ROI and how UK brands should run both in parallel. AI SEO and traditional SEO share vocabulary and almost nothing else. The mechanics, the signals and the outcomes are categorically distinct. This article is the side-by-side reference: where they overlap, where they diverge, and how UK brands should resource each in 2026.\n\n## The headline difference\n\nTraditional SEO optimises for placement in a list of ten links read by a human. AI SEO optimises for inclusion in a single synthesised answer written by a language model. The decision-maker, the format and the measurement are different in every respect that matters.\n\n## Side-by-side comparison\n\n- **Goal**: SEO ranks you in a list. AI SEO names you inside the answer.\n- **Decision-maker**: SEO targets a human evaluator. AI SEO targets a language model.\n- **Unit of value**: SEO measures clicks. AI SEO measures recommendations.\n- **Primary signals**: SEO weights backlinks, keywords and click-through rate. AI SEO weights entity clarity, coherence and third-party proof.\n- **Format**: SEO produces persuasion pages for human attention. AI SEO produces machine-legible knowledge for model inference.\n- **Measurement**: SEO measures position in a SERP. AI SEO measures probability of inclusion in an AI response.\n- **Time horizon**: SEO results compound over six to eighteen months. AI SEO produces first citations in fourteen days, category presence in sixty.\n- **Failure mode**: SEO failures are visible in analytics. AI SEO failures are silent: the demand is diverted before it reaches your funnel.\n\n## Where they overlap\n\nThe overlap is real but narrow. Both disciplines benefit from clean technical foundations, valid schema, fast page performance and high-quality writing. A site that is technically broken will underperform in both.\n\nBeyond that the disciplines diverge sharply. The work that wins SEO does not move AI inclusion, and the work that wins AI inclusion is, in many cases, invisible to traditional SEO measurement.\n\n## How to resource each in 2026\n\n- Categories where buyers still primarily Google: weight investment toward SEO, run AI SEO as a hedge.\n- Categories where buyers research with AI before contact: weight investment toward AI SEO, run SEO as the residual surface.\n- All categories: monitor your AI inclusion probability quarterly. The shift from one weighting to the other is happening faster than most teams budget for.\n\n## What does not work\n\nDoing more SEO does not solve AI exclusion. The signals are different. Adding schema to existing pages helps marginally; rewriting the entity, coherence and authority of the brand changes the outcome.\n\nBuying citations from low-authority directories is now actively harmful: it adds coherence noise without adding model-trusted corroboration.\n\n## How to start\n\nRun the diagnostic. Our free 60-second AI Recommendation Readiness Report returns your AI inclusion probability decomposed across the four AVO dimensions. From there the path is sequenced and measurable.\n\nThe disciplines are different. Treat them differently. Resource them both. --- ### ChatGPT vs Perplexity for B2B research: where your buyers actually are URL: https://avo-digital.co.uk/insights/chatgpt-vs-perplexity-for-b2b-research Category: Real-World Tests · Published: 12 Apr 2026 > We tracked 200 B2B buyer journeys across ChatGPT and Perplexity. The split is not what most marketing teams assume. Here is where to invest in 2026. ChatGPT and Perplexity are routinely treated as interchangeable AI search surfaces. They are not. The user populations, query patterns and selection mechanisms are different enough that B2B brands should resource them separately.\n\nThis article summarises what we have observed across 200 B2B buyer journeys tracked between January and March 2026.\n\n## The user populations are different\n\n- **ChatGPT**: dominant by raw query volume. Used across the entire buyer organisation, from first awareness to vendor shortlist.\n- **Perplexity**: lower volume but higher intent. Over-indexed on researchers, analysts, consultants, in-house counsel and senior decision-makers building defensible cases.\n\nIf your buyer is a senior, research-led professional, Perplexity is disproportionately important relative to its share of overall query volume.\n\n## The query patterns are different\n\nChatGPT queries lean conversational and exploratory: "what should I look for in a law firm for a series A", "how do I evaluate B2B SaaS vendors". Perplexity queries lean research-led and citation-hungry: "compare the leading B2B fintech compliance providers in the UK", "what is the regulatory position on X".\n\nThe practical implication: optimisation for ChatGPT and optimisation for Perplexity require different content shapes even when the underlying entity work is the same.\n\n## The selection mechanisms are different\n\n- **ChatGPT**: synthesises from training data plus optional retrieval. Citations are not always surfaced; selection is opaque.\n- **Perplexity**: surfaces five to ten citations per answer, weighted and visible. Selection is measurable.\n\nThis means Perplexity SEO is directly verifiable: you are in the citation set or you are not. ChatGPT requires inference: you are named more or less often, with growing confidence over time.\n\n## Where to invest in 2026\n\nFor B2B brands targeting senior, research-led buyers, the pragmatic split is:\n\n- Run AVO across both surfaces in parallel; the underlying entity, coherence and authority work compounds across each.\n- Tilt content investment toward Perplexity-shaped formats (deep, citation-hungry, structured) for the high-intent surface.\n- Tilt conversational and shortlist-shaped content (FAQ, comparison, "best X" formats) toward ChatGPT.\n\n## What does not work\n\n- Treating "AI SEO" as a single channel and producing one set of content for both. The query shapes do not match.\n- Ignoring Perplexity because the volume is small. The decision-makers it carries are not.\n- Optimising for one platform and assuming the other will follow. The selection mechanisms are different enough that gains do not always transfer cleanly.\n\nThe honest answer to "where should I invest" is "both, asymmetrically, with the asymmetry driven by who your actual buyer is". Most teams under-invest in Perplexity by an order of magnitude relative to its strategic value. --- ### Why Google AI Overviews change everything for UK SEO URL: https://avo-digital.co.uk/insights/why-google-ai-overviews-change-everything Category: AI vs Search · Published: 10 Apr 2026 > Google's AI Overviews now appear above traditional results on a growing share of commercial queries. What this means for ranking, traffic and the work that actually wins. Google's AI Overviews now appear above the traditional ten blue links on a rapidly-growing share of commercial UK queries. The Overview frequently answers the question entirely, citing a small handful of sources. If your brand is not in those citations, the click never happens, and your ranking below the Overview is increasingly cosmetic.\n\nThis is not a future scenario. It is happening now, and it changes the work UK SEO teams should be doing in 2026.\n\n## What an AI Overview actually is\n\nAn AI Overview is a synthesised answer generated by Gemini using Google's existing index plus the model's broader training. It selects from a far smaller set of sources than the traditional results, weighted toward entity clarity, structured data, freshness and third-party authority rather than conventional ranking signals.\n\nThe practical effect: a page can rank in Google's top three on a query and be entirely excluded from the AI Overview that sits above it.\n\n## The traffic implications\n\nWhere an AI Overview answers the query satisfactorily, click-through to the underlying results drops materially. Recent independent measurement puts the reduction at 20 to 40 percent on affected queries, depending on category and intent.\n\nThe businesses cited inside the Overview do not see this loss; if anything, their relative share grows. The businesses not cited absorb the entire decline, even when their traditional ranking holds.\n\n## What signals move AI Overview inclusion\n\nThe four AVO dimensions move inclusion in AI Overviews as much as they do in standalone Gemini, ChatGPT, Perplexity or Claude. Specifically:\n\n- **Interpretability**: tight entity definition, schema and structured data the AI Overview can reliably parse\n- **Coherence**: agreement across web, Google profile, directories and citations\n- **Authority**: third-party corroboration in sources Google's index already trusts\n- **Intent alignment**: content that answers the specific question end-to-end, not generic landing pages\n\n## What does not move it\n\n- More backlinks of any quality\n- More content of the same shape\n- More keyword targeting\n- Faster page load (a hygiene factor, not a lever)\n\nThe instinct to "double down on SEO fundamentals" is not wrong, but it is insufficient. The Overview's selection mechanism responds to a different signal set, and only investing in that set produces inclusion.\n\n## What to do now\n\n- Audit your current Overview inclusion across your top twenty commercial queries. Run our free 60-second audit.\n- Identify the queries where you rank in the top five but are not cited in the Overview. These are the highest-leverage targets.\n- Sequence the AVO work by dimension. Interpretability first, then coherence, then authority, then intent alignment.\n- Re-measure at sixty days. Inclusion in the Overview is binary and visible; the data is fast.\n\nGoogle's AI Overviews are not a tweak to SEO. They are the surface where the contest is increasingly being decided. Resource the work accordingly. --- ### Schema markup for AI search: the 2026 UK implementation guide URL: https://avo-digital.co.uk/insights/schema-markup-for-ai-search Category: AI vs Search · Published: 16 Apr 2026 > Schema is no longer just a Google rich-result lever. It is now the primary machine-legibility layer that ChatGPT, Gemini, Perplexity and Claude rely on to understand who you are. For a decade, schema markup was treated as a Google rich-result tactic: add Product, Review or FAQ schema, get a star rating in the SERP, move on. In 2026 the calculus is different. Schema is now the primary structured-data layer that large language models use to disambiguate your business when they generate answers and recommendations.\n\nThis is the practical UK guide to schema markup for AI search: which types matter, how to implement them, and the specific mistakes that make AI systems mis-categorise or omit you.\n\n## Why schema matters more for AI than for Google\n\nGoogle has twenty years of behavioural and link signals to fall back on when your schema is messy. Language models do not. When ChatGPT, Gemini, Perplexity or Claude attempt to resolve "who is this business and what do they do", schema is often the cleanest signal available, and it is weighted accordingly.\n\nA business with crisp, validated schema is materially more likely to be named in an AI answer than a competitor with stronger backlinks but no machine-legible structure.\n\n## The schema types that actually move AI inclusion\n\nFocus on these. The rest are noise.\n\n- **Organization** (or LocalBusiness for location-based services): the canonical entity record. Must include legal name, URL, logo, sameAs links to social and directory profiles, address, telephone, founding date.\n- **Service**: one Service entity per distinct offering. Avoid bundling everything into one record. Each Service should have a name, description, areaServed and provider.\n- **FAQPage**: directly consumable by AI systems building answers. Use real buyer questions in the buyer's exact phrasing.\n- **Article**: for thought-leadership content. Include author with credentials, datePublished, dateModified, publisher.\n- **Person**: for named experts and authors. AI systems weight content authored by identifiable, credentialled humans more heavily.\n- **Review** and **AggregateRating**: only with verifiable, real reviews. Synthetic ratings now actively damage authority signals.\n\n## The five mistakes that cause AI mis-categorisation\n\n1. **Ambiguous Organization name**. Listing the trading name in schema while the rest of the web uses a different brand. The model cannot reconcile, so it omits.\n2. **Missing sameAs**. The sameAs property links your Organization to LinkedIn, Companies House, Crunchbase and authoritative directories. Without it, the model has no way to corroborate.\n3. **Service descriptions written for SEO not humans**. Keyword-stuffed Service descriptions get ignored. Write what you actually do, in the language a buyer would use.\n4. **FAQPage built from generated questions**. AI systems detect synthetic FAQ patterns and discount them. Use the questions your sales team is actually answering.\n5. **Schema that contradicts the on-page content**. A model reading the page and the schema together will detect the mismatch and lower confidence. Schema must mirror reality.\n\n## How to implement: the priority order\n\n- Week one: deploy a clean, validated Organization (or LocalBusiness) record on every page. Include sameAs to all credible profiles.\n- Week two: one Service record per offering, each on its own page. Real descriptions, real areaServed.\n- Week three: FAQPage on the highest-intent commercial pages, using real buyer questions.\n- Week four: Article schema with Person author records on every long-form piece of content.\n\n## How to validate\n\nUse Google's Rich Results Test plus Schema.org's validator. Pass both. Then check that the schema is reachable in the rendered HTML, not injected after JavaScript hydration in a way that some crawlers miss.\n\nThe AVO Visibility Index audits your schema coverage automatically as part of the Interpretability dimension. Run the free 60-second report to see your current coverage and the specific gaps that are costing you AI inclusion.\n\nSchema is no longer a Google tactic with a side effect for AI. It is the machine-legibility layer that AI systems use to decide whether you exist. Treat it accordingly. --- ### What is llms.txt? The new standard AI crawlers actually read URL: https://avo-digital.co.uk/insights/llms-txt-file-explained Category: AI vs Search · Published: 15 Apr 2026 > llms.txt is the emerging standard that tells AI crawlers what to read on your site. Here is what it is, why it matters for ChatGPT and Perplexity inclusion, and exactly how to deploy one. llms.txt is the emerging counterpart to robots.txt for the AI era. Where robots.txt tells search crawlers what they may or may not index, llms.txt tells large language models what content on your site is worth ingesting and how it is structured.\n\nIt is not a W3C standard yet, but adoption by serious AI-search-aware brands is now widespread enough that omitting one is a competitive disadvantage. This is the practical guide for UK businesses.\n\n## What llms.txt actually does\n\nllms.txt is a single markdown file at the root of your domain (yourdomain.co.uk/llms.txt). It contains:\n\n- A short, plain-English description of your business and what your site offers\n- A curated list of the most important URLs on your site, grouped by category\n- Optional pointers to canonical content (whitepapers, methodology pages, fact sheets)\n\nAI systems that crawl your site can use it as a sitemap weighted by importance. Crucially, retrieval-augmented systems (Perplexity, ChatGPT search, Gemini grounding) increasingly prioritise llms.txt-listed pages when they are scoping which parts of your site to ingest for an answer.\n\n## Why this matters for AI search\n\nWithout llms.txt, an AI crawler making a snap decision about which of your URLs to read is choosing from your sitemap, your nav structure and your internal links. That selection is noisy. With llms.txt you direct the model to the canonical, high-signal pages. The result is better disambiguation and a higher likelihood of accurate inclusion in answers.\n\nWe have seen first-citation latency drop from approximately 28 days to under 10 days for clients who added a well-structured llms.txt file alongside the broader AVO work.\n\n## What a good llms.txt looks like\n\n\ --- ### AI search for local business: how UK SMEs win in voice and assistant queries URL: https://avo-digital.co.uk/insights/ai-search-for-local-business-uk Category: AI vs Search · Published: 13 Apr 2026 > When a UK customer asks ChatGPT, Gemini or Siri 'best plumber near me', the answer is now a single recommendation. This is how local SMEs win that recommendation. Local search has changed shape twice in five years. First the Map Pack compressed ten organic results into three. Now AI assistants compress those three into one. When a UK customer asks ChatGPT, Gemini, Siri or Google Assistant for the best plumber, accountant, dentist or solicitor near them, the answer is increasingly a single named business.\n\nThis is the practical guide for UK local businesses on how to be that business.\n\n## Why local AI search is different from local SEO\n\nLocal SEO optimised your Google Business Profile and chased reviews. Local AI search asks a different question: across all the places your business is described on the internet, do those descriptions agree, and do they triangulate to the location and service the user is asking about?\n\nA business can have a five-star Google Business Profile and still be invisible to ChatGPT if the Companies House record, the Yell entry, the LinkedIn company page and the website itself describe the business in subtly different ways. The AI cannot reconcile, so it picks the competitor that is internally consistent.\n\n## The five signals AI assistants weight for local recommendations\n\n1. **Name, address, phone (NAP) consistency at scale.** Not just on Google Business Profile. Across Companies House, Yell, Yelp UK, FreeIndex, Bing Places, Apple Maps, sector-specific directories and your own website. Even a punctuation difference (limited vs Ltd) reduces match confidence.\n2. **Service-area clarity.** AI systems need to map "near me" to a coherent geography. List your specific catchment areas in plain language on your service pages. "We cover Manchester, Stockport, Salford and Trafford" is parseable. "Greater Manchester area" is not.\n3. **Real, verifiable reviews on independent platforms.** Trustpilot, Google, Feefo, TripAdvisor where relevant. Volume matters less than verifiability and recency.\n4. **Schema for LocalBusiness.** Specifically with areaServed, openingHours, priceRange and a sameAs property linking to your Google Business Profile, LinkedIn and Companies House URL.\n5. **Editorial mentions in geographically relevant publications.** A mention in the Manchester Evening News carries more local weight than a generic national directory listing.\n\n## What does not work for local AI search\n\n- Stuffing service-area pages with neighbourhood names. Detected as low-quality and discounted.\n- Buying reviews. AI systems now triangulate review velocity, language patterns and reviewer history. Synthetic patterns reduce trust scores.\n- Generic blog content about your industry with no local specificity. Adds noise without signal.\n- Listing in 200 directories. After the top 15-20, additional listings add coherence noise.\n\n## The minimum viable local AI visibility deployment\n\nIf you do nothing else this month:\n\n- Audit your NAP across the top 15 UK directories. Fix every inconsistency.\n- Deploy LocalBusiness schema with areaServed listing your real catchment.\n- Add a service-area page per major town with genuine, distinct content.\n- Cross-link your Google Business Profile, LinkedIn, Companies House and any sector body listings via sameAs.\n- Solicit recent, verifiable reviews on at least two independent platforms.\n\n## Voice queries: the same signals, slightly different shape\n\nVoice queries to Siri, Alexa and Google Assistant resolve through similar selection mechanisms but tend to over-weight three things: review recency (last 90 days), proximity confidence (precise catchment definition) and openingHours (currently open vs not). If you are competing for "open now" voice queries, accurate openingHours schema is non-negotiable.\n\n## What to expect in 60 days\n\nLocal AI visibility moves faster than national. We routinely see UK SMEs go from zero AI assistant inclusion to first-position recommendation in their primary catchment within 45-60 days of completing the work above.\n\nRun the free 60-second AVO Visibility Index audit to see where you currently stand on each of the four dimensions, and which gaps are actually costing you local AI recommendations. --- ### ChatGPT Search ranking factors: what we know in April 2026 URL: https://avo-digital.co.uk/insights/chatgpt-search-ranking-factors Category: Real-World Tests · Published: 11 Apr 2026 > ChatGPT Search now drives material referral traffic to UK sites. Based on six months of controlled testing, here are the ranking factors that actually move inclusion. ChatGPT Search has graduated from novelty to a measurable referral source for UK businesses. Across the AVO research practice we now see it driving 4-12% of total organic-equivalent traffic for clients in research-led B2B categories, and trending up monthly.\n\nThe question every UK marketing team is asking: what actually moves a brand into ChatGPT Search citations? After six months of controlled testing across 200+ commercial queries, this is what the data supports.\n\n## How ChatGPT Search selects sources\n\nUnlike pure ChatGPT (which synthesises from training data), ChatGPT Search performs a live retrieval step against the open web, then synthesises. The retrieval step uses Bing's index as the primary substrate, filtered and re-ranked by OpenAI's own selection model.\n\nThis means two layers of ranking matter: getting indexed by Bing in the first place, and being selected by the OpenAI re-ranker for inclusion in the answer.\n\n## The signals that move ChatGPT Search inclusion\n\nIn order of observed impact:\n\n1. **Entity clarity in schema and on-page copy.** The single largest controllable lever. Brands that are precisely defined and consistently described across the indexed web are selected far more often than ambiguous ones, regardless of relative authority.\n2. **Topical depth on the specific query.** A 1,800-word article that genuinely answers the question outperforms a 4,000-word generic piece that mentions the topic in passing. The re-ranker rewards on-topic density.\n3. **Recency for time-sensitive queries.** Articles updated within the last 90 days are weighted significantly higher on commercial and "best of" queries. Stale evergreen content is being deprioritised.\n4. **Third-party corroboration.** Brands cited in two or more independent, indexed sources for a claim are selected more often than brands making the same claim in isolation. This is the AVO authority dimension in action.\n5. **Page-level structure.** Clear H2/H3 hierarchy, scannable lists, FAQPage schema and tight introductory paragraphs that summarise the article all increase selection probability. The re-ranker ingests structured pages more confidently.\n6. **Domain trust in Bing.** Domains that perform well in Bing's traditional ranking are over-represented in ChatGPT Search citations. The Bing index is the substrate, so Bing SEO is now an indirect ChatGPT Search lever.\n\n## What does not appear to move it\n\nThings the SEO industry sometimes assumes matter that we have not been able to correlate with ChatGPT Search inclusion in controlled tests:\n\n- Backlink volume in isolation\n- Social signals\n- Page load speed beyond a baseline of 3 seconds\n- Brand search volume (helps for direct queries, not for unbranded "best X" queries)\n- Adding more pages targeting the query\n- Internal linking density\n\n## The three-step UK playbook\n\n1. **Get indexed by Bing properly.** Submit your sitemap to Bing Webmaster Tools. Verify there are no robots-level blocks. Many UK SEO programmes have neglected Bing for years; this is now actively costing you ChatGPT Search inclusion.\n2. **Audit your top 20 commercial queries.** For each, identify whether your brand is currently cited in ChatGPT Search. The free AVO Visibility Index audit returns this for the AI engines we test.\n3. **Sequence the AVO work by dimension.** Interpretability and intent alignment first (highest leverage on ChatGPT Search specifically), then coherence and authority for compounding effect across all AI surfaces.\n\n## Time-to-result\n\nWe have observed first ChatGPT Search citations within 14-21 days of completing the priority work, with category presence stabilising at 60-90 days. This is materially faster than traditional SEO compounding, and the measurement is direct (you are in the citation set or you are not).\n\nChatGPT Search is no longer a future channel. It is a current one, with a predictable signal set, a fast measurement loop and disproportionate strategic value because the businesses that win it now will be the trained-in defaults of the next model generation. --- ### How to rank on ChatGPT in 2026 (the actual playbook) URL: https://avo-digital.co.uk/insights/how-to-rank-on-chatgpt-2026 Category: Real-World Tests · Published: 18 Jun 2026 > Forget the keyword tricks. Ranking on ChatGPT in 2026 is about being the source the model trusts to cite. Here's the 5-step playbook we use on every UK client. "How do I rank on ChatGPT?" is now the single most-asked question in UK marketing leadership meetings. The honest answer is that "rank" is the wrong word — ChatGPT does not return a list. It returns one synthesised recommendation, and your brand is either inside that recommendation or it is not.\n\nThis is the playbook we run on every AVO engagement, in the order we run it.\n\n## Step 1: Establish what ChatGPT currently says about you\n\nBefore touching anything, document the baseline. Ask ChatGPT (both the free and paid tiers, both with and without Search enabled) the 20 buyer-shaped queries that should surface your brand. Screenshot every answer. Note three things for each: are you mentioned, what is said about you, and who is mentioned in your place.\n\nThis is the only honest starting point. Most UK brands skip it and end up optimising for problems they do not have.\n\n## Step 2: Fix your entity, before anything else\n\nChatGPT does not select brands. It selects entities — structured, disambiguated, machine-readable representations of brands. If your entity is fuzzy, no amount of content will fix you.\n\nA clean entity has four things: a single canonical name used consistently across the indexed web, a one-sentence category definition that appears on your homepage and in your Organization schema, third-party corroboration (Wikipedia, Crunchbase, industry directories) using the same name and category, and a knowledge-graph presence Google has already validated.\n\nFix this layer first. Everything downstream compounds on it.\n\n## Step 3: Build the answer artefacts\n\nFor each priority query, you need an artefact ChatGPT can confidently cite. The shape that wins is consistent: a tight 60-word summary at the top that directly answers the query, H2 sections covering the obvious sub-questions, a comparison table where applicable, an FAQ block with FAQPage schema, and a named author with reviewer credentials.\n\nLength is not the lever. Density is. A 1,500-word article that says one thing well outperforms a 4,000-word article that says many things vaguely.\n\n## Step 4: Earn third-party corroboration\n\nChatGPT weights claims it sees in two or more independent sources far higher than claims that appear only on your own domain. This is the AVO authority dimension and it is the single largest gap in most UK marketing programmes.\n\nThe fastest path: contributed articles in trade publications, podcast appearances with full transcripts, comparison and "best of" inclusions, and structured directory listings (G2, Capterra, Clutch in B2B; industry-specific directories in B2C). The goal is not link equity. It is corroboration of your category claim.\n\n## Step 5: Measure on the surface, not the proxy\n\nTraditional SEO measures ranking position. AI visibility measures inclusion in the answer. These are different metrics and conflating them is the most common mistake we see.\n\nSet up a weekly probe of your 20 priority queries across ChatGPT, Gemini, Perplexity and Claude. Track inclusion rate, citation context (positive, neutral, comparative), and competitor presence. Move the inclusion rate, not the keyword position.\n\n## Time-to-first-result\n\nIn UK engagements over the last six months, we have observed first ChatGPT inclusions within 21-30 days of completing the entity and answer-artefact work, with stable category presence at 60-90 days. This is faster than traditional SEO because the measurement is direct and the surface is smaller.\n\n## What not to do\n\nDo not buy "AI SEO" packages that are repackaged blog content. Do not stuff "as an AI language model" into your copy. Do not chase llms.txt as a silver bullet (it helps; it is not the lever). Do not optimise for one engine in isolation — the dimensions that move ChatGPT also move Gemini, Perplexity and Claude.\n\nRanking on ChatGPT is an engineering problem, not a content problem. The brands that win are the ones who treat it that way. --- ### Best AI SEO tools UK [2026 review]: what we actually use URL: https://avo-digital.co.uk/insights/best-ai-seo-tools-uk-2026 Category: AI vs Search · Published: 16 Jun 2026 > We tested 14 AI SEO tools across six UK client engagements. Here's what genuinely moves AI visibility, what wastes budget, and the stack we run on every account. The AI SEO tool market exploded in 2025. By mid-2026 there are over 80 vendors claiming to "optimise for ChatGPT". We tested 14 of the most-cited ones across six UK client engagements and tracked actual movement in AI inclusion rates over 90 days.\n\nThis is what works, what doesn't, and the stack we now run on every AVO account.\n\n## The category, honestly\n\nNo single tool gets you cited by ChatGPT. The category is genuinely useful for diagnosis, monitoring and content QA, but the work that moves visibility is structural — entity, schema, authority, intent — and structural work needs a person.\n\nWith that caveat in place, here is the stack ranked by actual contribution to measured outcomes.\n\n## Tier 1: tools that materially moved client results\n\n**AVO Visibility Index** (free). Live probe across ChatGPT, Gemini, Perplexity and Claude with the dimension breakdown. We use it on every audit because it returns the four-dimension score in 60 seconds and the priority actions are specific to the brand. Bias acknowledged — it's ours — but it remains the only free UK-focused live-probe tool we know of.\n\n**Profound** (~$500/mo). Continuous monitoring of brand mentions across the major LLMs with sentiment and competitor tracking. Worth it for brands with active programmes where weekly visibility movement matters. Less useful for diagnosis-only engagements.\n\n**Schema App** (~$300/mo). Enterprise schema management. Genuinely moves entity clarity at scale. Overkill for sites under 200 pages; essential above 2,000.\n\n## Tier 2: useful but narrower\n\n**Otterly.ai** (~$70/mo). Lightweight prompt tracking. Good entry point if budget is tight and you only need to monitor 50-100 prompts. Less depth than Profound but a quarter of the cost.\n\n**Peec AI** (~$200/mo). Strong on competitor sentiment analysis specifically. Useful if you're in a hotly-contested category and need to understand why competitors are being preferred.\n\n**Surfer SEO** (~$90/mo). Traditional SEO content tool, but the recent AI-overview features genuinely help with answer-artefact density. We use it for content briefing.\n\n## Tier 3: did not move the needle\n\nWe will not name-and-shame, but five of the 14 tools tested either produced no measurable change in AI inclusion rates, double-counted improvements that were actually from other work, or generated content that demonstrably hurt entity coherence. As a category to be wary of: tools that promise to "rewrite your site for AI" without touching schema or third-party corroboration.\n\n## The stack we actually run\n\nFor a typical UK B2B client at £3-5K/month:\n\n- AVO Visibility Index (free) — weekly probe and dimension scoring\n- Profound (£400/mo) — continuous LLM monitoring\n- Schema App or Schema.dev (£200-300/mo, depending on site size)\n- Bing Webmaster Tools (free) — non-negotiable for ChatGPT Search\n- Google Search Console (free) — for AI Overviews tracking\n- A senior practitioner doing the entity, authority and intent work the tools cannot do\n\nTotal tooling spend: £600-700/month. Total programme cost: the rest. The ratio matters. Programmes that flip this (heavy tooling, light human work) consistently underperform.\n\n## How to choose if you're starting from scratch\n\nRun the free AVO Visibility Index first. If the score is below 40, you do not have a tool problem — you have a structural problem and no tool will fix it. If the score is 40-70, add Profound or Otterly to monitor improvement. If the score is above 70, add Schema App to scale entity clarity across your catalogue.\n\nThe best AI SEO tool is the one that surfaces the right next action. Everything else is noise. --- ### How to get cited in Perplexity: the UK research guide URL: https://avo-digital.co.uk/insights/how-to-get-cited-in-perplexity Category: Real-World Tests · Published: 14 Jun 2026 > Perplexity is the most predictable AI engine to win citations on — if you understand its retrieval model. Here's what we learned from 800+ tracked queries. Perplexity is the AI engine UK marketers most underestimate. It has a fraction of ChatGPT's user base, but the users it does have are disproportionately high-intent (researchers, analysts, B2B buyers), and the citation model is the most transparent of any major AI engine. If you want to learn how AI search ranking works, optimising for Perplexity is the fastest classroom.\n\nThis is what 800+ tracked queries across UK client engagements have taught us.\n\n## How Perplexity actually retrieves sources\n\nPerplexity performs a live web search for every query (no training-data fallback), retrieves 5-15 candidate sources, and synthesises an answer that explicitly cites a subset of them. The citations are visible and clickable. This makes Perplexity uniquely measurable: you can directly see whether you were cited, and if so, in which position.\n\nThe retrieval substrate is a hybrid: Perplexity's own crawl, plus the Bing index for breadth. The selection model is more recency-weighted and more authority-weighted than ChatGPT Search.\n\n## The signals that move Perplexity citations\n\nIn order of observed impact in our test set:\n\n1. **Topical authority at the domain level.** Perplexity favours domains that have published consistently on a topic over time. A single great article on a generalist site loses to a moderate article on a specialist site. This is the most asymmetric signal in AI search and the one most UK brands ignore.\n2. **Recency.** Articles updated within 60 days are heavily favoured on commercial and current-events queries. We have watched citations shift within 48 hours of a republish.\n3. **Source diversity in your own corpus.** Perplexity de-duplicates by domain in many answer sets. Having multiple distinct pages targeting the query family (a pillar plus 4-6 supporting articles) increases the probability that at least one is cited.\n4. **External corroboration.** Same authority signal as ChatGPT, weighted slightly higher in Perplexity. Brands cited in trade publications and industry directories appear in Perplexity citations significantly more often.\n5. **Schema and structured data.** Less critical for Perplexity than for ChatGPT Search, but FAQPage and Article schema still help.\n6. **Backlinks.** The only AI engine where backlinks show a measurable correlation with citation rate. Not the dominant signal, but present.\n\n## The 4-week Perplexity playbook\n\n**Week 1: Probe.** Run your 30 priority queries through Perplexity. Document citation presence, position and competitor citations. The free AVO Visibility Index covers Perplexity in the default probe set.\n\n**Week 2: Publish or refresh.** For each query where you're absent, identify the strongest existing article on your site and update it with a 60-day-fresh date, a tighter summary paragraph, a comparison table where applicable, and 2-3 new sourced statistics. Where no article exists, publish one.\n\n**Week 3: Corroborate.** Pitch one guest article or podcast appearance per priority query family on a Perplexity-indexed publication. Trade press in your category is more valuable than mass-market press.\n\n**Week 4: Re-probe and iterate.** Re-run the queries. In our test set, 35-50% of refreshed articles gained citation within four weeks. The remainder typically need an authority intervention rather than a content intervention.\n\n## Why Perplexity matters strategically\n\nThree reasons. First, the measurement loop is fast enough to learn from. Second, Perplexity's user base over-indexes on buyers who write the briefs that hire you. Third, the signals that win Perplexity citations generalise. Brands that perform well in Perplexity tend to perform well in ChatGPT Search and Google AI Overviews within 2-3 months, because the underlying signals (topical depth, recency, corroboration) are shared.\n\nIf you want one AI engine to optimise for first, this is the one. --- ### AI SEO cost UK [2026]: what to actually budget URL: https://avo-digital.co.uk/insights/ai-seo-cost-uk-2026 Category: AI vs Search · Published: 12 Jun 2026 > Honest pricing for AI SEO and GEO in the UK in 2026. What £500/mo gets you, where £5K/mo is the right floor, and the red flags in cheap packages. "How much does AI SEO cost?" is the question every UK marketing director asks before the second meeting. The honest answer is "it depends on what you're actually buying", and the market is currently flooded with packages priced as though they were the same thing when they are very much not.\n\nThis is the price map, based on what we see across the UK GEO market in mid-2026 and what each tier actually delivers.\n\n## The four real tiers\n\n**Tier 1: DIY plus tooling. £100-300/month.** A few monitoring tools (Otterly, the free AVO Visibility Index, Bing Webmaster Tools) plus your existing team's time. Honest answer: this works for brands with strong in-house SEO and a real entity foundation. It will not work if your category is contested or your entity is unclear.\n\n**Tier 2: Productised AI SEO packages. £500-1,500/month.** Templated audits, recurring content drops, schema add-ons. Reasonable starting point for early-stage businesses and local services. Will move your visibility from invisible to baseline. Will not get you cited in competitive UK B2B categories.\n\n**Tier 3: Mid-market GEO programmes. £2,500-6,000/month.** Full four-dimension work: entity, interpretability, authority, intent. A named senior practitioner, weekly probes, custom artefact production, third-party corroboration outreach. This is where measurable category presence happens for most UK B2B brands. AVO's mid-market programmes sit in this band.\n\n**Tier 4: Enterprise and contested categories. £8,000-25,000/month.** Multi-market, multi-language, multi-brand. Dedicated team, custom monitoring, integration with paid search and PR. Required for finance, healthcare, legal, enterprise SaaS and any category where the top three brands are actively investing in AI visibility.\n\n## What changes the right tier for you\n\nFour variables move the answer:\n\n- **Category competition.** Local trades: Tier 2 is usually enough. UK B2B SaaS: Tier 3 is the floor. Regulated industries: Tier 4 or you lose.\n- **Entity baseline.** A brand with a clean Google knowledge panel and Wikipedia presence starts ahead. A brand with three name variants and no schema starts behind and needs 2-3 months of foundational work that's the same cost regardless of tier.\n- **Existing SEO maturity.** Strong technical SEO halves the time-to-first-citation. Weak technical SEO adds 2-3 months and £5-10K of remediation.\n- **In-house capacity.** Teams with senior content and SEO people can absorb a lot of the programme. Teams without need an agency to do more, which costs more.\n\n## Red flags in cheap packages\n\nThe under-£500/month "AI SEO" packages we have audited consistently include:\n\n- Generic blog content with no entity work and no schema\n- "AI optimisation" that is keyword stuffing with "ChatGPT" added to old SEO copy\n- llms.txt files as the headline deliverable (helpful, not the lever)\n- Monthly reports that measure traffic instead of AI citation rates\n- No baseline probe and no measurement of the actual outcome (AI inclusion)\n\nIf the package does not include a live probe across ChatGPT, Gemini, Perplexity and Claude, you cannot tell whether it is working. Walk away.\n\n## What honest pricing looks like\n\nFor a UK B2B with a £40-80M revenue and a contested category, the right floor is £3-5K/month for the first six months (foundational entity, authority and artefact work), then £2-3K/month ongoing for monitoring and iteration. Smaller businesses can run leaner; enterprise needs more.\n\nROI breakeven typically lands at month 4-6 in our engagements, because AI-sourced traffic converts at 2-3x the rate of paid traffic (the buyer has already been pre-qualified by the AI's recommendation). Once the engine is running, the cost per qualified lead is materially lower than paid search in most B2B categories.\n\n## How to scope your own budget\n\nRun the free AVO Visibility Index first. The dimension scores tell you which tier you actually need:\n\n- Score 0-30: structural problems. Tier 3 or above is the only honest answer.\n- Score 30-60: foundations partly in place. Tier 2 or Tier 3 depending on category.\n- Score 60-85: good baseline. Tier 2 plus targeted work on the weakest dimension.\n- Score 85+: optimisation mode. Tier 1 plus selective expert input.\n\nThe right budget is the one matched to the work you actually need, not the tier the vendor wants to sell you. --- ## Glossary ## Machine-readable data - AVO Visibility Index (JSON, Dataset schema): https://avo-digital.co.uk/visibility-index.json - AVO Visibility Index (CSV): https://avo-digital.co.uk/visibility-index.csv - Insights RSS feed: https://avo-digital.co.uk/feed.xml - Site index for AI: https://avo-digital.co.uk/llms.txt - Fact sheet (canonical entity reference): https://avo-digital.co.uk/fact-sheet