Entities and Pages and Companies, Oh My!
AI engines name different brands for the same query 62% of the time
Only 17% of queries return the same brands across all engines
Contradictory sources make buyers 66% less likely to close
Brand mentions predict AI visibility ~3x more than backlinks
B2B buyers now use about 10 channels, double 2016
AI does not decide what to say about your company by reading your website. It reads the whole web at once, looks for the version of you that every source agrees on, and repeats that. The work is no longer publishing the right page. It is becoming a fact the internet corroborates.
Ask ChatGPT, Google’s AI Overviews, and Perplexity the same buying question and you will get three different shortlists. BrightEdge, running tens of thousands of identical prompts across the major engines, found they disagreed on which brands to recommend 61.9% of the time, and named the same set across all of them in only 17% of cases. For any given question, the machine that introduces you to a buyer is effectively picking from a different memory than the one next to it.
That disagreement is the tell. It means these systems are not looking up a fact about you. They are reconstructing you, on the spot, from whatever the web collectively seems to say. When the web is consistent, they reconstruct you confidently and name you. When it is not, they hedge, and hedging means naming someone easier to be sure about. The question that decides your visibility is no longer whether your page ranks. It is whether the internet tells one coherent story about who you are.
The unit of visibility changed
For twenty years the unit of SEO was the page. You optimized a URL, earned links to it, and moved it up a ranking. An answer engine does not work in pages. It works in entities: the single, resolved concept of your company that it assembles from every source at once, your site, your LinkedIn, a directory, a review platform, a podcast transcript, a comparison article you have never seen. The page is one input. The entity is the thing that gets named.
This is why a beautifully built website can be invisible in AI answers while a plainer competitor gets cited constantly. The competitor is not winning on pages. It is winning because the web agrees on what it is, and agreement is what a probabilistic system reaches for when it has to decide whom to name. You are not competing to rank. You are competing to be the version of your category the machine is most sure about.
Entity consistency
The machine can’t agree on who you are. Neither can your buyers’ sources.
Ask three AI engines the same question and most of the time they name different brands. Buyers checking a supplier’s story run into the same problem.
How often AI engines agree
If your story changes between sources, the machine trusts none of them, and names someone else.
BrightEdge compared brand recommendations for identical prompts across ChatGPT, Google AI Overviews and Google AI Mode. Bars use the full 0–100% scale, with a mark at 50%. Gartner and McKinsey figures describe B2B buyers.
Underneath the AI disagreement sits an older, human version of the same problem, and it is worth naming because it tells you the stakes are real and not a quirk of the models. Gartner found that when buyers encounter trustworthy but contradictory information across their research, they are 66% less likely to complete a significant purchase. Contradiction does not just confuse the machine. It stalls the human. And buyers are exposed to more of it than ever: McKinsey’s B2B Pulse research puts the typical purchase across about ten channels now, double the five it took in 2016. Ten surfaces are ten chances to describe yourself slightly differently, and every mismatch is a small subtraction from trust, first the model’s, then the buyer’s.
When your site says one thing and a directory says another, the machine does not pick the flattering version. It lowers its confidence in both, and names the company it is surer about.
Why a file about yourself proves nothing
The instinct, when you learn that machines read your site, is to write the machine a better document: a crisp about page, a page of claims, a file that states plainly how good you are. It feels like the obvious move, and it is largely wasted. A page you wrote about yourself is a single, self-interested source. It proves you are willing to say the thing. It does not prove the thing is true, and a system trained to weigh corroboration treats a lone self-claim accordingly.
What moves the needle is agreement you do not fully control. Ahrefs, in a study of roughly 75,000 brands, found that how often a brand is mentioned across the web correlates with its visibility in AI answers about three times more strongly than backlinks do. Not links you built. Mentions you earned, on pages you do not own, describing you the same way your own does. That convergence, the same name, the same category, the same claim, showing up across independent sources, is what tells a model you are a real, resolved thing safe to put in an answer. It is the difference between saying you exist and the web confirming it.
AI visibility
What earns the mention is proof. Not files.
A footprint the web agrees on beats anything you write about yourself.
What correlates with being named in AI Overviews
Mentions correlate far more strongly with AI Overview visibility than backlinks do.
A file you wrote proves you say it.
A footprint the web agrees on proves it’s true.
Being an entity is slow. That’s exactly why it’s a moat.
Figures are correlation coefficients between each factor and brand visibility in Google AI Overviews, on a scale from 0 to 1. Bars use that full scale, with a mark at 0.5. Correlation shows association, not cause.
There is a hard rule hiding in that chart, and it is the whole discipline in one line: what you assert about yourself is weak evidence, and what others independently repeat about you is strong evidence. Which reframes the work. You are not writing copy for a crawler. You are trying to get the same true, specific description of your company to appear, unprompted, in enough trusted places that a machine reading all of them at once concludes there is only one version of you.
The three places a story fractures
Incoherence is rarely dramatic. It is small drift, accumulated across surfaces nobody owns end to end. Three places account for most of it:
The name and the mark. The same company rendered as three slightly different names, an old logo here, a dead domain there. A machine cannot be sure two near-matches are one entity, so it discounts both.
The claim. What you do, and who you are for, stated one way on your site and a different way on a directory or an old profile. The model cannot resolve which is your real category, so it stops confidently placing you in either.
The corroboration. The description on the sources you do not own. If the third-party web is silent or stale about you, there is nothing for the model to check your own claims against, so it treats them as unverified.
None of these is a content problem in the way marketing usually means it. They are consistency problems, and consistency is unglamorous, cross-functional, and owned by no one, which is exactly why fixing it is an edge. The reconciliation nobody wants to do is the work that compounds.
Why this is a moat, not a hack
Here is the part that should change how you budget. If a single file or tag could buy AI visibility, everyone would deploy it within a week and it would be worth nothing. The reason an entity works is that it cannot be bought in a quarter. It accrues: a mention here, a corroborated fact there, a description that finally matches across ten surfaces, built up until the web’s account of you is dense and consistent enough that no competitor can dislodge it with a burst of content. Slow to build is the same as hard to copy. That is the definition of a moat.
And it is durable in a way tactics are not. Ranking factors shift, features get deprecated, this quarter’s trick stops working. But a coherent, well-corroborated entity is what every current answer engine reads and every future one will too, because “who does the independent record say this is” is not a Google preference. It is how any system reasons about trust. You are not optimizing for the current model. You are building the thing the next several models will all rely on, while most competitors are still writing themselves another page.
How to start today
Ask the machine who you are. Put your top buyer questions to ChatGPT, Perplexity, and Google’s AI Mode and record how each describes you, what it cites, and where it confuses you with someone else. Disagreement between engines is your map of where your entity is unresolved.
Fix the name and the mark first. Make your company name, primary domain, and logo identical everywhere they appear, starting with your own site, your LinkedIn, and the directories that rank for your name. A near-match reads as a different company.
Reconcile the claim across every surface. Write one plain, specific description of what you do and who you are for, then make the sources you control match it word for word, and correct the ones you can influence.
Earn one corroborating mention you do not own. A podcast, a guest piece, an analyst note, an accurate directory listing. One independent source that describes you the way you describe yourself is worth more than a page of self-claims.
Keep score on the entity, not the page. Track whether the assistants name you, describe you accurately, and agree with each other over time. A falling disagreement rate is the signal the web is converging on one version of you, which is the version that gets cited.
Sources
BrightEdge, via Search Engine Land, “Google AI, ChatGPT rarely agree on brand recommendations” (Aug 2025) — engines disagreed on brand recommendations 61.9% of the time; only 17% of queries returned the same brands across all three. https://searchengineland.com/google-ai-chatgpt-brand-recommendations-data-461423
Gartner — buyers who encounter trustworthy but contradictory information are 66% less likely to complete a significant deal; buyers spend only ~17% of the journey with suppliers. https://www.gartner.com/en/newsroom/press-releases/2019-07-29-gartner-reveals-new-b2b-sales-approach-to-win-in-toda
Ahrefs, 75,000-brand study — brand web mentions correlate with AI visibility ~0.66 versus ~0.22 for backlinks, roughly 3x stronger. https://searchengineland.com/used-cited-brands-appear-ai-search-481664
McKinsey, B2B Pulse — B2B buyers now use about ten interaction channels in a purchase, up from five in 2016. https://www.mckinsey.com/capabilities/growth-marketing-and-sales/our-insights/b2b-sales-omnichannel-everywhere-every-time