Which sources should you fix first if AI is not naming you?
Summary
Fix in the order of how much cited material each surface accounts for, and fix mechanical problems before editorial ones. Yext's October 2025 analysis of 6.8 million citations across 1.6 million AI answers found 86% came from sources a brand can manage or influence — 44% first-party websites, 42% listings and profiles, 8% reviews and social, 6% news and forums. So the sequence is: confirm a crawler can read the page at all, resolve which company you are, bring existing profiles into line with one canonical description, then rewrite the pages buyers land on. Mass directory submission is the specific failure this data argues against.
Start with the surfaces you can edit today, in the order of how much of the cited material they account for. Yext analysed 6.8 million citations from 1.6 million AI answers across Gemini, OpenAI and Perplexity in October 2025 and found 86% came from sources a brand can manage or influence — 44% first-party websites, 42% listings and profiles, 8% reviews and social, and 6% news, forums and other uncontrolled sources. So the honest ranking is: your own site first, your listings a very close second, and almost everything else after that.
The instinct is usually to write more content. The data says the second-largest cited surface is a set of profile pages most companies have not opened since somebody created them.
Why do listings account for so much of what gets cited?
Because they answer the question directly and in a consistent format. A directory entry states what a company does, where it is, and what category it belongs to, in a structure that is trivially retrievable. A homepage often states a value proposition, which is harder to extract a fact from.
There is a second reason that matters more. Engines cross-check. When your site, your professional profile and three directory entries all describe you the same way, that agreement is itself evidence. When they disagree — an old address here, a different company description there, a category that no longer fits — the entity becomes less certain, and an uncertain entity is one a system is less willing to name in a recommendation. Inconsistency does not just fail to help. It actively subtracts.
What is the right order to fix things in?
Mechanical problems before editorial ones, and identity before content. Doing it in the other order means improving pages that either cannot be read or cannot be attributed to you.
- Can a crawler read the page at all? Vercel and MERJ tracked over 500 million GPTBot requests in December 2024 and found no evidence of JavaScript ever being executed, with GPTBot requesting JavaScript files 11.5% of the time and ClaudeBot 23.8% and neither running them. If your content only exists after scripts run, nothing below this line matters yet.
- Is it unambiguous which company you are? One canonical description, one consistent legal and trading name, structured data declaring the organisation with a stable identifier, and a complete set of real profile links. If your name collides with another business, resolve that before publishing anything else, or every improvement strengthens the wrong entity.
- Do your existing profiles agree with each other? This is the listings and profiles surface Yext measured at 42%. Audit what exists before creating anything new.
- Do your own pages answer the questions buyers ask? This is the first-party surface Yext measured at 44%, and the one you have complete control over.
- Is anyone else describing you, and accurately? The remaining 14% in the Yext split — reviews, social, news and forums. Worth knowing about; not worth chasing first.
How do I audit listings without wasting a week?
Write the canonical description first — one paragraph, the version you would want quoted verbatim — then compare every surface against it rather than editing each one on its own merits. Without that reference text you will produce five more slightly different descriptions, which is the failure this exercise exists to fix.
| Field | The failure to look for |
|---|---|
| Company name | Legal entity on one profile, trading name on another, an abbreviation on a third |
| Description | Three generations of positioning coexisting across three platforms |
| Category | A category chosen when the business did something slightly different |
| Address and contact | A previous office, or a phone number nobody answers |
| Website URL | Pointing at http, a redirect chain, or a page that no longer exists |
| Ownership | A profile nobody at the company can currently log in to |
That last row is the one that stalls this work most often. Find out who controls each profile before you plan the rewrite, not after.
Should I submit my company to as many directories as possible?
No. The Yext finding describes where citations came from; it says nothing about the return on adding another low-quality listing. Mass submission is an efficient way to manufacture inconsistent descriptions at scale, which is precisely the failure the cross-checking argument warns about. Fix the surfaces that already exist, then add only those that are genuinely relevant to your industry and that a buyer might plausibly read.
Does this replace working on my website?
No — in the same Yext analysis, first-party sites were still the single largest cited surface at 44%. The point of the ordering is sequencing, not substitution. Listings come earlier than most people expect because they are quick, cheap and currently wrong; the website is a larger and slower job that you will still be doing afterwards.
What the split does argue against is a content-only strategy. If you publish twenty new pages and leave five contradictory profiles in place, you have improved the 44% Yext attributed to first-party sites while continuing to undermine the identity that all of it has to attach to.
What can a company realistically change this quarter?
All of the following are weeks of work, not quarters, and none of them require a website rebuild:
- Run the crawler-visibility test on three page templates and fix it if it fails. Days, usually.
- Write the canonical description. An afternoon, and it unblocks everything else.
- Inventory every profile that exists, including the ones nobody remembers creating, and record who controls each. A day.
- Bring every profile into line with the canonical description. A week, spread out.
- Rewrite the six pages a buyer would actually land on, answer-first, with question-shaped headings. Two to three weeks.
- Write down the buyer questions you want to be found for, run them across engines, and keep the results as a baseline. Half a day, and without it you cannot tell whether any of the above worked.
Do the last one first, even though it is listed last. A baseline taken after the fixes is not a baseline.
Where to start
If you have not yet seen which sources the engines currently cite when they answer a question in your category, that is the first hour of work and it is not one you should pay for. The work after that is paid, and I publish what an audit and the implementation behind it cost, so it can be weighed against doing the same sequence in-house. The service that follows is getting found when buyers ask AI, and the underlying research is covered in half of what AI cites about you is not on your website.