How ChatGPT decides which businesses to recommend

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.

The selection mechanism is specific, learnable and engineerable. This article documents what we have observed across twelve months of controlled testing.

The five mechanisms in priority order

ChatGPT's recommendations emerge from the interaction of five underlying mechanisms. They are not weighted equally.

• Entity legibility: can the model classify what the business does, for whom, where and at what stage of buyer journey, with high confidence
• Cross-platform coherence: do the signals about the business agree across the open web (website, Google profile, directories, social, citations)
• Third-party authority: how often is the business referenced by sources the model already trusts (industry press, professional bodies, trade publications)
• Question alignment: how well does the business's content answer the actual question the user asked, end to end
• Recency and freshness: when was the supporting evidence last updated, and is it consistent with the model's most recent training cut

Mechanism one: entity legibility

Below 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.

Legibility 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.

Mechanism two: coherence

Once 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.

In our tests, resolving coherence conflicts produced an average +14 point uplift in inclusion probability with no other intervention.

Mechanism three: authority

Authority 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.

Authority 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.

Mechanism four: question alignment

ChatGPT'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.

This is why generic landing pages underperform and why a single deeply-answered FAQ frequently outperforms an entire library of SEO articles.

Mechanism five: recency

ChatGPT's training cut, browsing tools and retrieval-augmented features all favour recently-updated, demonstrably-current sources. Stale evidence is downweighted.

This does not mean publishing more. It means visibly maintaining the evidence that already exists, so the model has a current corroboration to draw on.

What this means in practice

If 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.

Most teams attempt all five in parallel and dilute the result. Sequence is the durable insight here.