What 5,000 website scans tell us about AI search

We scanned 4,999 US business homepages between 11 and 13 October 2026, using the same free scan anyone can run on GeoTest. The median GEO Score is 49 out of 100. Not one site earned an A. Most sites can be read by AI crawlers without trouble; what they lack is anything an AI engine can quote: a direct answer at the top of each section, and specific facts it can repeat with your name beside them.

Key facts from the benchmark:

• 4,999 homepages scanned and published, scored under GEO page score v6.1.
• The median score is 49 and the mean is 47.5. The lowest is 8 and the highest is 80.
• No homepage scored 90 or more (grade A). One scored 80 (grade B). 7.9% scored 65 or more.
• 53.1% scored below 50, which is grade E or F.
• 96.8% of homepages have sections that do not open with the answer.
• 91.5% have too few concrete facts. 88.6% miss both of these checks.
• Home services businesses had the highest median of the local groups (53); community groups had the lowest (41).
• Sites built on AI app builders had a median of 36. On 88.8% of Lovable sites the main text needs JavaScript to appear.

How was the benchmark built?

The benchmark is 4,999 homepages, each scanned once with the GeoTest single-page scan between 11 and 13 October 2026. Sites were chosen by trade and by city, never by score, and every scan is listed on the public benchmark board.

We left out directories, marketplaces, publishers and public bodies, because they are not ordinary businesses: 17 such pages were removed by hand after the run. Sites whose robots.txt blocks all crawlers, and scans that failed to load, were skipped.

Every site got the scan a visitor gets on the GeoTest homepage: one page, the homepage, scored on the published rubric. On a single page the score also reflects how much of the rubric the page gave the scan to test, so a page with almost nothing on it cannot score well by default.

What does a typical business homepage score?

A typical business homepage scores 49 out of 100, which is grade E (Poor). More than half the sample falls in grades E and F, and the grades that tell an owner the page is in good shape (A and B) hold a single site between them.

The ceiling is low as well as the middle. Only 123 homepages (2.5%) reached 70, and 13 reached 75.

Which kinds of business score best and worst?

Movers, online stores and home services businesses scored highest, and community groups and sites made with AI app builders scored lowest. The spread between the best and worst group is 21 points of median score.

Health and legal sites carry an extra rule: a site that gives health, legal or financial advice is capped at 79 if its homepage has no Privacy or Terms link, because those are the pages a reader and an engine use to check who stands behind the advice.

What is missing on most homepages?

The two checks most homepages fail are both about being quotable: sections that open with the answer, and specific facts. 99.6% of homepages fail at least one of the two.

An answer-first section is one whose first sentence states the point in full, so an engine can lift it without reading on. Most homepages open each section with a slogan or a welcome line instead, and 456 homepages (9.1%) have no subheadings at all, so there is no section for an answer to open.

Fact density is the rate of checkable specifics: a price, a year founded, a service area, a response time, a number of jobs done. "Fast, friendly, reliable service" gives an engine nothing to repeat; "Same-day repairs across Fairfax County since 2009" does.

The last row is worth a second look. 5.1% of these homepages, about one in twenty, tell search engines not to index them at all, usually a setting left on from when the site was built. The GEO Score caps a noindexed page at 30, because a page search engines are told to skip cannot be retrieved for an answer.

Where do these sites lose their points?

These sites lose most of their points on what they say, not on whether a crawler can read them. Scored as a share of what each category could have earned, the technical categories are mostly fine and the content categories are mostly empty.

Freshness is left out of this table because a homepage carries no publication date to score; it applies to articles, documentation and product pages.

That split is the main finding of the benchmark. Website builders, hosting and themes have largely solved the technical layer for small businesses. Writing the page so an AI engine can quote it has not been solved by any of them, and that is now where the score is decided.

Does a longer homepage score higher?

Longer homepages scored higher, but not because length is rewarded: the GEO Score gives no points for word count. A longer page simply has more room for the things that are scored, such as sections that answer a question, specific facts and a questions block.

The median homepage has 813 words a crawler can read. Adding words without adding facts or answers will not move the score; the 300-word homepage that states its prices, area and hours will beat the 1,500-word one that does not.

Why do AI-built websites score lower?

Sites made with AI app builders scored lowest because on most of them the text is drawn by JavaScript after the page loads, so a crawler that reads the raw page receives an almost empty shell. That was true of 88.8% of the 233 Lovable sites, 92.9% of the 56 Base44 sites and 73.9% of the 23 Replit sites in the sample.

Smaller platform groups (Wix 22, Framer 48, Bolt 15, v0 4 sites) are on the benchmark board but too small to rank here. The fix for an AI-built site is not a rebuild: prerendering the pages, or moving to a framework that renders on the server, puts the text in the raw page where every crawler can read it.

What should a business fix first?

A business should fix the two quotability checks first, because nearly every homepage fails them and they sit in the two categories where these sites earned the fewest points.

1. Open every section with the answer. Under each heading, make the first sentence state the point in full: what you do, where, for whom, and at what price or speed.
2. Replace adjectives with facts. Years in business, service area, prices or price ranges, response times, licences, number of customers. Put each one in its own sentence.
3. Add five to eight real customer questions with direct answers, and mark them up with FAQPage schema so the questions on the page and in the code match.
4. Link a privacy policy and terms from the homepage. Two out of three homepages in this sample do not, and health, legal and finance sites are capped at 79 without them.
5. Check the page with JavaScript turned off. If the text disappears, fix that before anything else, because every other improvement is invisible to a crawler that cannot read the page.

What does this benchmark not show?

This benchmark measures how ready a homepage is to be quoted by AI engines; it does not count how often any of these sites is actually cited. Citation depends on the question asked, the competition for it and each engine's own retrieval, which a page scan cannot see.

• Homepages only. A business whose best content lives on service pages or a blog will score higher on a full site analysis than its homepage does here.
• US only, found through search. Local businesses were found by searching trades in US cities, so businesses that do not appear in search results are under-represented.
• One rubric version. Every score is GEO page score v6.1. Reports scored under earlier versions use different rules and are not comparable with these numbers.

Frequently asked questions

What is the average GEO Score of a small-business website?

The median homepage in the GeoTest benchmark scores 49 out of 100 and the mean is 47.5, across 4,999 US business homepages scanned in October 2026. Just over half (53.1%) score below 50, and none reached the 90 needed for an A.

What is the most common reason a homepage scores low?

Sections that do not open with a direct answer, and too few concrete facts. 96.8% of homepages missed the first and 91.5% missed the second; 88.6% missed both. Those two checks sit in the categories where the benchmark sites earned the lowest share of their points.

Do websites built with AI app builders score lower?

In this benchmark, yes. The 331 sites built on AI app builders had a median of 36, against 53 for sites on Webflow, Wix, Squarespace and Framer and 55 for Shopify and WooCommerce stores. On 88.8% of Lovable sites the main text only appears after JavaScript runs, so a crawler that reads the raw page sees a median of 277 words.

Can I see where my own site ranks against the benchmark?

Yes. Run a free scan on the GeoTest homepage. It is the same single-page scan used for every site in the benchmark, scored by the same rubric version, and the public benchmark board lists every published scan.

Does a longer homepage score higher?

Longer homepages scored higher in this sample (a median of 56 above 1,500 words against 38 under 300), but length earns no points on its own. Longer pages simply carry more of what does: sections that answer a question, specific facts, and a visible Q&A block.

Free GEO resources: The GeoTest Benchmark Report 2026 and the AI Search Readiness Checklist