Insights — Digital Infrastructure — 3 min read
How Do AI Systems Decide Which Companies to Recommend?
AI systems do not use a published ranking formula the way search engines once did. They lean on entity recognition, source corroboration and retrieval quality — and understanding these mechanics, honestly, matters more than chasing rumoured tricks.
In short
AI systems generally combine three elements when recommending a company: whether the model or its retrieval layer can find and confidently identify the company as a distinct, well-defined entity; whether independent sources corroborate the same facts about it; and whether its own content directly and clearly answers the question being asked. Beyond that, each platform has its own undisclosed weighting, and it changes over time — so specific tactics beyond these fundamentals should be treated as unverified.
When a business asks why a competitor gets mentioned by an AI assistant and it does not, the underlying question is really about mechanics: what is the system actually doing when it decides which companies to name in an answer. There is no single, universal mechanism, because different tools — a chatbot with live search, a chatbot answering purely from training data, an AI-generated summary at the top of a search results page — work differently, and none of the companies operating them publish a full ranking formula.
What can be said with reasonable confidence is drawn from how these systems are publicly known to be built: they combine large language models with retrieval methods that find and rank source material, and increasingly with entity databases that try to resolve who or what a company actually is. This article sets out what is known, what is plausible but unconfirmed, and what stays genuinely uncertain, rather than presenting invented ranking factors as fact.
Entity recognition: being a clearly defined 'thing' the system can identify
Modern AI systems increasingly reason in terms of entities — a specific company, product or person — rather than pure keyword matching. For a business to be recommended, a system generally needs to first resolve that the business is a real, distinct, identifiable entity, with a consistent name, description and set of facts attached to it. A company whose name, service description or location varies across its own website and third-party listings is harder to resolve confidently, and an ambiguous entity is a weak candidate for a confident recommendation.
Corroboration: does more than one source agree?
A single unverifiable claim on a company's own website carries less weight than the same claim appearing consistently across the company's site, independent directories, industry press and customer reviews. This mirrors how human due diligence works and is consistent with published descriptions of how retrieval-augmented systems weigh source reliability: agreement across independent sources increases confidence; a lone unverified claim does not.
Retrieval quality: can the system actually find and extract a clear answer?
Even a genuinely well-regarded company will be passed over if its content does not present facts in a form a retrieval system can extract cleanly — plain, self-contained statements rather than answers buried in narrative or spread across several pages with no single clear passage. This is a mechanical, structural factor, separate from the company's actual quality or reputation.
Signals that plausibly matter, and how confident to be in each
| Signal | How confident should a business be that it matters? |
|---|---|
| Clear, consistent entity description | Reasonably confident — consistent with how entity resolution is known to work |
| Third-party corroboration (reviews, press, directories) | Reasonably confident, exact weighting unknown |
| Structured data / schema markup | Plausible, unconfirmed weighting |
| Direct, extractable answers to specific questions | Reasonably confident — matches how retrieval-augmented generation is documented to work |
| Backlink volume alone | Weak evidence it transfers directly from traditional SEO |
| Paid placement | Not currently how most AI assistants work, though this is evolving and platform-specific |
Where genuine uncertainty remains
What a company can reasonably do
- Make the company an unambiguous, consistently described entity across its own site and third-party sources
- Build genuine third-party corroboration — reviews, press mentions, directory listings — rather than relying solely on self-description
- Write content that answers specific buyer questions in clear, extractable passages
- Use structured data to state facts explicitly rather than leaving them to be inferred
- Treat this as an ongoing discipline, not a one-off project, given how quickly the platforms themselves change
Why this matters commercially
For B2B companies, being recommended or omitted at the research stage of a buying decision increasingly happens before a human ever visits a website. Understanding the honest mechanics — rather than chasing rumoured tricks — is what lets a business invest sensibly in the parts of this it can actually influence: clarity, consistency and corroboration.
Could your commercial operation run with less admin and better information?
Evans Sales Consultancy applies AI, automation and practical digital systems to prospecting, sales operations, reporting, customer journeys and management visibility — starting from the commercial problem, not the technology.
Related services