Half of what AI cites about you is not on your website

The finding

In October 2025 Yext published an analysis of 6.8 million citations drawn from 1.6 million AI-generated answers across Gemini, OpenAI and Perplexity. It found that 86% of citations came from sources a brand can directly manage or strongly influence, split 44% first-party websites and 42% listings and controllable profiles, with reviews and social accounting for 8% and news, forums and everything else for the remaining 6%.

Why the listings share is the uncomfortable number

Most of the effort in this discipline goes into the first-party 44% that Yext measured. Restructuring pages, writing self-contained passages, fixing structured data — that is the work people mean when they say answer engine optimisation, and it is real work with real effect.

The 42% that Yext attributed to listings and profiles tends to get audited once, during onboarding, and then never again. Directory entries written by someone who left three years ago. An industry association listing with a superseded product line. A profile whose one-line description contradicts the positioning on the homepage. None of it is visible in your analytics, none of it appears in a site crawl, and all of it is being read.

There is a compounding problem underneath this. Engines cross-check across sources. When your own site and three directories describe the business differently, the effect is not that one description wins — it is that the entity itself becomes less certain, and an uncertain entity is an easy one to leave out of a shortlist.

What to audit, in priority order

1. Your own canonical description. Before touching anything external, write the single description of the business that everything else has to match. One paragraph. If you cannot produce it in one sitting, that is the finding, and no amount of profile editing fixes it.
2. Search your company name in each engine. Ask ChatGPT, Perplexity, Gemini and Claude "what is [company]" and look at what gets cited. Those citations are your actual entity footprint, as opposed to the one you assume you have.
3. Industry and trade directories. Usually the largest cluster in B2B, and usually the most out of date. Check the description, the category, the contact details and the URL.
4. Professional profiles. Company and personal profiles on professional networks. The one-line descriptor is doing more work than its length suggests.
5. Third-party mentions you do not control. Articles, roundups, partner pages. You cannot edit these, but knowing what they say tells you what an engine has been told.

The consistency test

Put every description you found in one document, side by side. Then check three things: does each one describe the same business, do they use the same primary term for what you sell, and does any of them contradict another on a checkable fact.

The second is the one that quietly does the most damage. A company describing itself as a "precision components manufacturer" in one place, an "engineering solutions provider" in another and a "CNC machining specialist" in a third has given retrieval three different categories to place it in — and will lose to a competitor that used one term consistently, even if that competitor is smaller.

What this does not mean

It does not mean listings are more important than your website. In the Yext data the first-party share (44%) still edges the listings share (42%), and the site is where you control the passage structure that decides whether a citation is quotable. It also does not mean submitting to every directory that will take you: the Yext figure describes where citations came from, not the return on adding one more low-quality listing, and a mass-submission campaign is a good way to create the inconsistency described above at scale.

What it means is proportionality. If roughly half the cited surface area sits outside your content management system, spending all of the effort inside it is a decision worth making deliberately rather than by default.