56 well-known sites scored 81/100 for SEO and 43/100 for AI visibility

One of them fails six basic SEO checks and is named in every AI answer we asked for. Another passes almost everything and is named in none. Traditional search health and AI search visibility are not the same measurement, and this is what the gap looks like with the numbers attached.

The headline: a 38-point gap

These are large, well-run companies. Most have real SEO teams, and it shows: the median SEO score in this set is 86. Then we asked an answer engine the questions their buyers would actually type, and recorded which brands it named.

Only 4 of 56 (7%) scored 70 or above for AI visibility. 14 scored under 30. One was named in none of the six buying questions asked in its own category.

The two results that matter most

Averages hide the interesting part. Two sites in this set point in exactly opposite directions, and together they are the whole finding.

GoCardless: SEO 90, AI visibility 15, named in 0 of 6

Technically excellent. It passes nearly everything a traditional audit looks at. And when we asked an answer engine the payment questions its buyers ask, it was named zero times. Stripe was named in 100% of those answers, PayPal in 83%. A direct competitor owns the category inside the answer, and no amount of on-page work had prevented that.

Wise: SEO 71, AI visibility 73, named in 6 of 6

The mirror image, and the more useful one. Wise scores 71 out of 100 on technical SEO, near the bottom of the set. Our scan found its page set to noindex, its content only appearing after JavaScript runs, no valid sitemap, no llms.txt, no Organization schema and a missing meta description. On a traditional audit it would be handed a long list of failures.

It was named in every single answer we asked.

That single result kills the comfortable assumption that AI visibility is just SEO with a new name. If it were, Wise would be invisible. It is not, because the rest of the web talks about it constantly, and an answer engine is reading the rest of the web.

What this suggests

Being cited by an answer engine appears to depend far more on what other websites say about you than on how tidy your own is. That is consistent with the widely reported finding that around 85% of the sources AI engines cite are third-party pages rather than the brand's own site, and it matches the pattern here: the sites with the biggest gaps are strong technically and quiet elsewhere.

The practical read: technical work sets your ceiling, and off-site presence decides where you sit under it. You need both, but they are not interchangeable, and most businesses are currently only doing one.

The full results

Sorted by the size of the gap between SEO and AI visibility. A positive gap means the site is stronger on traditional SEO than it is inside AI answers.

Every row links to its own full report from the reports index, including the exact questions asked and which brands were named instead.

Method

• Sample: 56 websites, measured 2026-10-02. Chosen for recognisability across several categories (developer tools, design, project management, UK fintech, retail), not randomly sampled.
• AI visibility: six buying questions were generated for each site's own category, grounded in live web search, then asked of a web-grounded answer engine. We recorded which brands it named. The score blends that outcome with a readiness measure of how quotable the site's own pages are.
• SEO: deterministic fetches and HTML parsing across 14 signals: indexability, JavaScript dependence, titles, canonicals, duplicate hosts, sitemaps, headings and so on. No model judgement.
• Exclusions: a site was left out entirely rather than published with a number we did not trust. Any check we could not measure was dropped from the score rather than counted as a failure.

Limitations, stated plainly

This is a small study and we would rather say so than have someone else say it for us.

• 56 sites is a small sample. Treat the individual results as solid and the averages as indicative.
• One measurement, one day. AI answers vary between runs and change over time. A rescan next month would not produce identical numbers.
• One answer engine, web-grounded, not four. We did not separately poll ChatGPT, Claude, Perplexity and Gemini for this study. Since those engines overlap only around 11% in the domains they cite, a per-engine study would likely show meaningful variation, and that is the obvious next piece of work.
• The sample is chosen, not random. These are recognisable brands, which biases the set toward companies with the resources to do SEO well. If anything that makes the gap more striking, not less.
• Six questions per site. Enough to be indicative of a category, not exhaustive.
• We sell a product in this space, which you should factor in. That is also why the raw per-site data is published rather than summarised: you can check any row.

Writing about this? Cite it freely, no permission needed. Press and data has the figures with their methods, our own score including the half that is bad, and a standing offer: tell us the domains your piece covers and we will scan them free and send you the results, whether or not you mention us.

Reusing this

The data is free to cite with a link to this page. If you want a cut we have not published, or want your own site measured, the scan is free and takes about a minute.

Questions, answered