Buyers are already working this way. Gartner’s survey of 645 B2B buyers found that 45% used generative AI during a recent purchase, mainly to gather information on vendors and products.
Ask an assistant to recommend suppliers in a category and it will name a few companies. Ask again with slightly different wording and the list changes a little, but not completely. Some companies appear reliably. Most never appear at all.
The question worth understanding is what separates them. Not the mechanics of any particular system, which change, but the properties that make a company mentionable.
In this article
The problem the system is solving
A system producing a recommendation has a specific difficulty. It has to name a company without being able to verify anything itself.
That makes it risk-averse in a particular way. Naming a company that turns out not to exist, or that is described wrongly, is a failure. Declining to name anyone is a lesser failure. So the bias is toward companies that are unambiguously real, consistently described, and corroborated by sources that are independent of the company itself.
That single observation explains most of what follows. The question is not “who is best”. It is “who can be named with confidence”.
A system naming suppliers is managing risk. It favours companies that resolve to one unambiguous entity, described consistently in sources the company does not control.
Being one entity, not several
Before a system can mention you, it has to establish that you are a single, identifiable organisation.
This sounds trivial and frequently is not. Companies routinely appear as several near-identical entities. A trading name and a legal name used interchangeably, an address written four ways, two phone number formats, a founding year on one page and nowhere else, a professional profile describing a different service mix from the website.
To a human, these are the same company. To a system resolving entities, each inconsistency reduces confidence that the mentions refer to one organisation.
The shared vocabulary for declaring these facts explicitly, rather than leaving them to be inferred, is published at schema.org and is worth using for exactly this reason.
The fix is otherwise unglamorous and entirely within your control. Decide the canonical form of every fact about your organisation. Use it identically everywhere, without variation, and update every location when anything changes.
For a company operating from a place buyers may not know well, this matters more. A studio in Thiruvananthapuram, Kerala, benefits from every source stating the same city, the same address, the same country, in the same form, because the system has less ambient knowledge to fall back on than it would for a company in a heavily documented location.
Corroboration from sources you do not own
Your own website is a claim. Independent sources are evidence.
A company described consistently across its own site, industry directories, professional networks, client sites, news coverage and review platforms is far easier to name with confidence than one that exists only on its own domain.
This is why the least glamorous marketing activities have become disproportionately valuable. A client publishing a case study about working with you. A trade publication quoting your founder. An accurate directory listing. A conference programme with your name on it. Each is an independent record.
The pattern to aim for is simple to state and slow to build. The same facts about you, appearing in places you do not control, agreeing with each other.
Factual density
Between two pages on the same subject, the one containing more checkable facts is more useful to a system producing an answer, because an answer needs substance to be made of.
Factual density means specifics. Numbers with baselines, dates, named entities, defined terms, described methods, stated conditions.
Consider two descriptions of the same company.
“A leading digital agency delivering innovative solutions for businesses worldwide.”
“A B2B web design and development studio in Thiruvananthapuram, India, incorporated in 2017 and continuing a practice begun in 2005, with more than 1,000 websites delivered across more than 40 countries.”
The first contains no facts. Nothing in it can be checked, quoted or used to answer a question. The second contains six checkable claims.
Matsio publishes that 85% of its clients return for further work, alongside how it was measured, across every engagement since 2005. The basis is attached deliberately. A figure with its basis can be evaluated, while a figure without one can only be repeated or ignored.
Most B2B websites are dense with adjectives and thin on facts. The adjectives are invisible to the process being described here.
Category association
Being known is not enough. You have to be known for something specific.
Systems answering “who does X” are looking for organisations associated with X, and that association is built by repeatedly appearing in contexts about X.
The practical consequence is that breadth hurts. A company that describes itself as doing web design, branding, marketing, video, events and software development has a diffuse association with all of them and a strong association with none. A company that describes itself narrowly, and publishes consistently within that narrow area, builds a clear one.
This is the same conclusion positioning has always reached. What is new is that the cost of being unfocused is now paid at the retrieval stage, before any human judgement is involved.
Geography works the same way. A company that wants to be associated with a place has to be described in connection with that place, consistently, in sources beyond its own site.
What does not seem to help
Some tactics are promoted for this and are either ineffective or harmful.
Volume without substance. Publishing many thin pages on adjacent keywords produces a large, low-density record. It gives a system more text and not more facts.
Claims of leadership. “Award-winning”, “leading”, “top-rated” are unverifiable and universal. They occupy the space where a checkable fact could be.
Manufactured mentions. Listings and placements bought at volume are typically identifiable as a group and are not independent in any meaningful sense.
Contradicting yourself for different audiences. Describing your company one way to one market and differently to another creates exactly the ambiguity that suppresses mentions.
Currency, and the cost of stale records
Systems assessing whether to name a company are also assessing whether what they know is still true.
Stale signals are common and easy to fix. A page describing a service you no longer offer. A team listing people who left. A copyright year two years old. A case study that was current when written and now describes a version of your business that no longer exists.
Each of these is a small contradiction with your current description, and contradictions reduce confidence.
The habit worth adopting is a scheduled review of the factual surface of your business rather than of its marketing. Once or twice a year, check every page that states a fact, every external profile and every directory entry, and correct what has drifted. It is an afternoon and it is the highest-yield maintenance available.
Being honest about what is known here
A caution, because this area attracts more confident advice than it can support.
The mechanisms by which any particular system selects companies are not public, they differ between systems, and they change. Anyone offering a precise formula, or guaranteeing a mention, is describing something they cannot know.
What can be said with reasonable confidence is structural rather than mechanical. Systems that produce answers must resolve entities, must prefer corroborated information over uncorroborated, and must have factual material to construct an answer from. Those constraints follow from the task, not from any implementation.
So the advice in this article is deliberately limited to properties that follow from the nature of the problem. They are also, usefully, the properties that made a company credible to human researchers before any of this existed, which means the investment is not contingent on a particular technology continuing to work in a particular way.
What to do, in order
Fix your facts first. One canonical form of every organisational detail, used everywhere. This is a day of work and it is the foundation for everything else.
Write the plain description. One sentence stating what you do, for whom, where, since when, at what scale. Put it on your site, in every profile, in every directory, unchanged.
Increase factual density. Go through your main pages and count the checkable claims. Replace adjectives with facts wherever the fact exists.
Earn independent mentions. Client case studies published by clients. Contributions to industry publications. Accurate listings. Speaking. Slow, and the most durable part.
Narrow the association. Decide what you want to be retrieved for and publish consistently within it.
Check what is said about you. Ask several assistants what your company does, periodically. Inaccuracies point to a source that is thin, stale or inconsistent, and that source is usually findable.
Related reading. how AI assistants read websites covers parsing, how structured, factual writing earns citations covers the writing, and what generative engines reward draws the three properties together. For how buyers use these answers, see how buyers research vendors through AI chat.
A note on the geography of this
One consequence is worth drawing out for companies outside the heavily documented markets.
Systems have more ambient information about companies in places that are extensively written about, and less about everywhere else. A company in a widely covered city inherits some benefit of the doubt. The location is unambiguous, the business environment is documented, and adjacent references are plentiful.
A company in a less documented location does not inherit that, which means more of the work has to be done explicitly. Stating the city, region and country in full, in the same form, everywhere. Ensuring the address resolves and matches registration records. Building independent mentions that connect the company to the place.
This is not a disadvantage so much as a difference in what has to be made explicit. The companies that handle it treat their location as a fact to be stated precisely and often, rather than as a detail to be mentioned once on a contact page.
The short version
A system naming companies is managing risk. It favours organisations that resolve to one unambiguous entity, are described consistently in sources they do not own, carry a high density of checkable facts, and are associated with a specific category.
None of those are tricks, and none can be acquired quickly. That is what makes them worth building.
A small thing and a big thing
One small thing to fix on your website today, and one big thing to learn that gets you more leads.
Your information is
safe with us. No spam.