Your SEO Is Fine. Your AI Visibility Needs Its Own Measurement.

Why Can a Site Rank Well and Still Be Misdescribed by AI?

A site can rank well and still be misdescribed by AI because ranking and answering are different jobs. A search ranking places your page in a list of links. An AI assistant writes an answer, and that answer can name you, leave you out, name you without a link, or describe a different business that shares your name. None of those outcomes shows up in a rank report.

We have seen both failures firsthand. In November 2025, Perplexity described Pixelmojo as not offering two services we did offer. In July 2026, a customer audit found three of four engines describing a different, same-named business. Search Console had nothing to say about either.

What Do SEO Tools Measure?

SEO tools measure how a search engine sees your site: where your pages rank for keywords, how often they are shown and clicked, who links to you, and whether search crawlers can reach and index your pages. Those numbers answer one question well: can someone searching find a link to your page?

The left column is mature. Search Console gives you impressions and clicks for free, and every SEO suite tracks rankings and backlinks. The right column is newer, and it is where the gap opens: a page that ranks first can still be absent from, or wrong in, the answer an assistant writes about the same topic.

What Does AI Visibility Measurement Need?

AI visibility measurement needs four kinds of evidence: whether AI bots can reach your pages, whether answers name you, whether answers link you, and whether they describe the right business correctly. Each one is a separate observation, and a good result in one does not imply the others.

Access

Access is whether a bot can fetch the page. It depends on two things that often disagree: what your robots.txt allows for each bot, and what your server or firewall actually does when that bot asks. A policy can allow a bot that the firewall then blocks. Checking only the policy misses that.

Mention and citation

A mention is an answer that names your brand. A citation, in Radar's reporting, is an answer that links one of your pages as a source. The two are counted separately because an answer can name you without linking you, and readers who never click still see the name.

Accuracy and identity

Being named is not the same as being described correctly. An answer can name you and still attribute the wrong services to you, or merge you with another business that has the same name. The wrong-company audit in Part 6 is what happens when a tool skips this step.

Do SEO Suites Measure AI Visibility Now?

Some do. An earlier version of this post said no SEO tool measures AI visibility. That stopped being true: as of 26 September 2026, both Semrush and Ahrefs sell AI answer tracking, and most dedicated AI visibility tools also check crawler access.

The prices and features come from our comparison of AI visibility tools, which records the date each one was checked. If you already pay for Semrush or Ahrefs, check what your plan includes before buying anything else.

The useful distinction is not SEO tool versus AI tool. It is tracking versus diagnosis. A tracking subscription watches a set of prompts over time. A diagnostic audit checks one domain in depth at one moment and tells you what to look at first. Many teams need both, and some need neither yet.

Why Do AI Answers Need Repeated Sampling?

AI answers need repeated sampling because the same question can get a different answer on the next run. A September 2026 study of product queries found that the products recommended often changed across repeated requests. One screenshot of one answer is an anecdote, not a measurement.

A measurement you can defend has three properties:

• Several prompts. One question tells you about one question.
• Several engines. ChatGPT, Claude, Gemini and Perplexity do not agree with each other, and readers use all of them.
• Counts beside every rate. "Named in 3 of 8 answers" can be checked. "37% visibility" cannot, unless the counts are shown.

In Radar, the Details tab shows the prompts, separate per-engine mention and link results for answers that name your brand, and the counts behind every rate. We set out the full standard in The AI Visibility Evidence Standard.

Does llms.txt or Schema Close the Gap?

There is no evidence that llms.txt or extra schema changes what AI engines say about you. Google says Google Search ignores llms.txt, and no major AI engine documents using it to choose citations, as of September 2026. Schema is still worth getting right, but as accuracy and hygiene, not as a citation lever.

The SE Ranking study covers about 300,000 domains. Google's guide to its generative AI features, published in May 2026, says its AI features need no special files or markup. We go through both in Google Says You Don't Need llms.txt.

Radar still checks llms.txt and schema in the free check. A missing or broken file is cheap to fix and worth knowing about, and some coding agents do fetch llms.txt. Radar shows that evidence note beside the result so the check is not mistaken for a ranking factor.

How Do You Measure Your Own Gap?

Measure the gap in four steps, from free to paid, and stop at the step that answers your question. Most teams start with data they already have.

1. Your search baseline. Search Console shows which pages earn impressions and clicks. That is your SEO picture, and you probably have it already.
2. A free technical check. Radar's free check tests AI crawler access, robots.txt, llms.txt, schema markup, AEO page readiness and an overall readiness score. It is one check per domain and needs no card. An AI model writes a short summary of the results, but the check does not ask AI engines about you.
3. What the engines say. A paid audit runs all 13 tools, including Reddit monitoring and the tools that sample answers from ChatGPT, Claude, Gemini and Perplexity. Read mentions and links separately, and check that the answers describe the right business.
4. A recheck after a fix. A recheck shows what changed. It does not show why, because the engines also change on their own.

Read the transcript

The site loads. But what does AI say? A crawl check helps inspect access. A sampled answer shows what a model returned to a question. Those are different observations. Start with the client's question. Check access when that's the concern. Inspect the answer to understand the description. I'm Scout, your guide to Radar.

What Changed in This Post?

We rewrote this post on 2 October 2026 as part of a review of every post on the blog. The changes:

• the claim that no SEO tool measures AI visibility is removed, because Semrush and Ahrefs now sell AI answer tracking;
• statistics we could not trace to a linked source are removed, including a Fortune 1000 schema figure, an AI referral share range and a conversion comparison;
• llms.txt, schema and formatting are no longer described as levers that change AI answers;
• the list of layers now matches what Radar measures, and the tool list matches its 13 tools.

What This Series Covers

This is Part 3 of The AI Visibility Stack, a six-part series on how we built Radar and what we learned running it. Part 2 tells the origin story from the git history. Part 4 covers how one audit reads 13 tools together.