State of GEO: October 2026

Read this first: why October starts a new series

Engine 4.19.0, released on 9 October 2026, made rubric v2 the default. Rubric v2 has nine categories instead of eight and adds a Google AI readiness category worth 20 of the 100 points. Every category's maximum changed, so a v2 total and a v1 total do not measure the same thing. The June to September editions were scored on v1 and are not placed next to these figures.

• Window and cohort. 767 unique domains, from user-initiated audits recorded between 9 and 31 October 2026 (UTC) and scored on rubric v2. 197 further domains whose only October audits ran before 4.19.0, on v1, are left out.
• Category detail covers 518 of the 767 domains. The benchmark table began storing the Google AI readiness score on 19 October. Audits recorded before that carry a valid v2 total but no stored value for that category, so the category table below uses the 518 domains audited from 19 October onward. Every other figure uses all 767.
• Categories do not add up to the total. For those 518 domains the nine category averages sum to 69.57, while the average total is 67.15. The engine subtracts 1, 3 or 5 points when it detects negative signals, and that deduction is not stored per category. The 2.42-point average difference, never above 5, is consistent with it; the benchmark does not record which signal triggered it.
• Self-selected sample. These are sites that someone chose to audit with GeoReady, a group likely to care about AI visibility more than the average site. Figures should not be read as rates for the web.

Score distribution: most sites sit in Good or Foundation

The average GEO score across 767 unique domains is 66.2 out of 100 on rubric v2, not comparable with the rubric v1 editions, and the median is 69. The lowest score is 14 and the highest 98. The bottom tenth of sites scored 45 or less, the bottom quarter 59 or less, the top quarter 76 or more and the top tenth 82 or more.

434 of 767 sites (56.6%) reached Good or better, and the other 333 (43.4%) sit in Foundation or Critical. Only 41 sites (5.3%) reached Excellent. The band ranges are the ones published on the methodology page, and every band in this cohort sits inside them (lowest Good score 68, lowest Foundation 36).

Category breakdown: one category adds almost the same points to every site

The table shows each of the nine rubric v2 categories for the 518 domains with category detail: the average points, the maximum, the share of the maximum the average site earns, the quartiles and the share of sites that score zero in the category.

Google AI readiness averages 17.75 of 20 points. Its quartiles are 18, 18 and 19, and the lowest score in the 518 domains is 8. A category that almost every site passes adds about the same number of points to each total, so it does little to separate strong sites from weak ones. It is also one reason a rubric v2 total cannot be set against a rubric v1 total: the maxima of the other categories changed too. Meta tags (92.7%) and content quality (78.9%) are the other categories most sites do well.

Among the 518 domains with category detail, the categories that separate sites are llms.txt, where the median is 5 of 6 points and 38.2% of sites score zero, schema (17.6% at zero, quartiles 7 to 11) and brand and entity (38.5% of the maximum). AI discovery is last at 12.3% of its 3 points, with 73.6% of sites at zero.

Each bar is the share of the 767 domains where the audit detected the signal. They are detection rates from one fetch of the home page and its standard files, not a count of every page on a site.

The basic on-page signals (language, canonical, H1, a direct opening) are present on more than four sites in five. The AI-specific ones are not: a full llms.txt on 26.1%, an AI discovery endpoint on 15.4%. The average site's robots.txt explicitly allows 24.4 AI crawlers, and the average page has 1,431 words.

llms.txt: widely present, often only a stub

GeoReady (geoready.dev) detected an llms.txt file on 476 of 767 domains (62.1%). 200 of them (26.1% of all domains) serve a full, structured file. The other 276, 58% of the sites that have a file, serve a file the audit does not count as a full one, which in practice is a short placeholder.

The 62.1% is a floor. When a CDN or firewall answers the auditor with a 403 or 406, the audit retries once with a browser-like user agent. A file that is still refused counts as absent, and the benchmark does not record how often that happened.

Why this rate is far above the top-site scans

Other published scans of llms.txt report much lower rates. As read on 1 November 2026, Rankability's tracker (18 September 2026) reports 8.3% of the Tranco top 10,000 domains (a domain counts if either llms.txt or llms-full.txt is confirmed), and SEOmator's scan (13 September 2026) reports 8.6% of 100,000 domains, or 11.4% of the 75,353 that responded. These figures do not conflict with ours. They measure ranked lists of domains, while this cohort is 767 sites whose owners chose to run an audit, and each source decides in its own way what counts as a valid file. None of the three is the rate for the web.

llms.txt and schema together

Crossing the two most common structured signals gives four groups. The second column shows each group's share of the cohort and the third its average total score.

55.4% of sites have both and average 75.1; 12.9% have neither and average 41.4. The mixed groups differ: sites with schema but no llms.txt average 64.5, while sites with llms.txt but no schema average 47.4, below the schema-only group. The score awards points for both signals, so these averages are partly built in. The next section removes those points.

Which signals go with a higher score, after removing their own points

The simple comparison of sites with and without a signal overstates the gap, because the rubric awards points for the signal itself. The table gives both versions: the raw average total score, and the average total with the points of the category that holds the signal subtracted. The second figure is the better description of how the rest of the site looks.

Removing the signal's own points shrinks the llms.txt, schema, FAQ, answer-first and AI discovery gaps, and most of each remains. The llms.txt gap goes from 15.5 to 10.4 points, the schema gap from 28.4 to 18.7, FAQ markup from 16.9 to 12 and an answer-first opening from 18.5 to 12.2. Sites with an AI discovery endpoint score 5.3 points higher on the rest of the rubric. The blocking gap does not shrink: sites that block an AI crawler score 54 against 56 on the rest of the rubric, 2 points lower, and 1.5 points lower on the raw total.

These are associations in a self-selected sample. Our reading is that a site that publishes one of these signals tends to publish others and to maintain its pages more carefully in general. Whatever the reason, the gaps do not show that adding a signal raises a score, and they say nothing about whether an AI engine cites a site more often.

AI crawler access: most sites block nothing, a few block eleven or more

9.8% of the 767 domains (75 sites) block at least one AI crawler in robots.txt, and 6% block three or more. The table shows how many crawlers each blocking site names.

25 of the 75 blocking sites (33.3%) block a single crawler, and 18 block eleven or more. Our reading is that the second group applies a blanket rule rather than choosing crawlers one by one, but the data cannot show intent. The audit stores how many crawlers a site blocks, not which ones, so this report names no individual crawler and gives no per-crawler percentage. For what each crawler does and how to set the rules, see the robots.txt guide for AI bots.

AI discovery: the category most sites skip

15.4% of the 767 domains expose at least one AI discovery endpoint, such as a /.well-known/ai.txt file. In the 518 domains with category detail the category averages 0.37 of 3 points, the median is 0 and 73.6% score zero. Sites with an endpoint average 70.4 on the rest of the rubric against 65.1 for the 649 without one, a gap of 5.3 points that, as above, is an association.

Engine versions in the cohort

Nine releases of the open-source engine scored the cohort. Each domain was scored by the version live on the day of its latest audit. 283 domains (36.9%) were scored on 4.19.8.

The averages by version are not estimates of what each version does to a score. Versions follow the calendar, so each one scored a different set of sites on different days, and four of them scored fewer than 50 domains. The detection changes in 4.19.1, 4.19.7 and 4.19.8 can raise a site's brand and entity points; their effect on this cohort was not measured. Every release is in the public changelog.

By domain suffix

Eleven suffixes have at least ten domains in the cohort. Groups of eleven to twenty-eight sites move a lot with a few additions, so read these as a description of this sample.

The largest group, .com, has 421 of the 767 domains and sits at 66.6, close to the cohort average. Among the groups with at least ten sites, .app has the highest average (70.4) and .io the lowest (59.2). llms.txt adoption runs from 38.5% (.in, 13 sites) to 87.5% (.app, 16 sites).

Where the points are

Based on where audited sites in this cohort lose the most points, these are the changes with the most room. Each one is a way to earn rubric points, and none of them is a promise of citations.

1. Replace a stub llms.txt with a complete file. 58% of the sites that have a file have no full one. A full file describes the site, groups the key pages in sections and links real URLs. The free llms.txt generator gives a starting structure.
2. Check that bots can fetch the file. A CDN or firewall that answers non-browser clients with a 403 or 406 can hide the file from the audit even after its browser-like retry, and possibly from other tools. Request it with a non-browser user agent.
3. Add Organization and WebSite schema if you are in the 31.2% without Organization markup. Include name, url, logo, sameAs and description. See Organization schema and sameAs.
4. Publish an AI discovery file. 84.6% of sites have none. It is the category where sites earn the smallest share of the available points: 12.3% on average across the 518 sites with category detail.
5. Read your robots.txt for blocks you did not intend. If you opt out of a crawler on purpose, check which crawlers the rule covers. 18 sites in this cohort block eleven or more.
6. Mark up real question-and-answer content, and open key pages with the answer. 27% of sites have FAQ markup and 83.1% open with a direct answer. See FAQPage schema for AI search.

Methodology

This report covers 767 unique domains audited by GeoReady (geoready.dev) between 9 October 2026 and 1 November 2026 00:00 UTC. When a domain was audited more than once, only its latest audit in the window counts, so each domain appears once, in the state of its last audit.

• Population filter: user-initiated audits recorded by the GeoReady web app (source = 'web') and scored on rubric v2. Scheduled monitoring re-checks are excluded. 197 domains whose only October audits used rubric v1 (before 9 October) are excluded.
• Engine: GEO Optimizer 4.19.0 to 4.19.8, the open-source engine behind GeoReady (geoready.dev). Rubric v2 has nine categories and 100 points. Maxima: Google AI readiness 20, robots.txt 14, schema 14, content 14, brand and entity 12, meta tags 11, llms.txt 6, technical signals 6, AI discovery 3. Source on GitHub; definitions on the methodology page.
• Band ranges: Excellent 86 to 100, Good 68 to 85, Foundation 36 to 67, Critical 0 to 35.
• Category table: 518 of the 767 domains, those whose audit stored the Google AI readiness score (recorded from 19 October 2026). Category averages therefore describe a different, smaller group than the headline figures, and they sum to 69.57 against an average total of 67.15 because of the negative-signal deduction described above.
• Adjusted averages in the signals table subtract, from each site's total, the points of the category that holds the signal (for Organization schema, schema and brand and entity together). Percentages are computed from counts and rounded to one decimal, so groups may sum to 99.9 or 100.1.
• Domain identity: a domain is identified by the last two labels of its hostname, so multi-part suffixes such as co.uk merge distinct sites into one row. The same rule applied in every edition. Domains are deduplicated by a salted HMAC-SHA256 hash, and only aggregates are published.
• What the data cannot say: it does not measure whether any AI engine cites a site, it holds the number of crawlers a site blocks and not which ones, and it covers audited sites rather than the web. No comparison with earlier editions is made because they used rubric v1.
• Third-party figures (Rankability and SEOmator, in the llms.txt section) were read on their own pages on 1 November 2026 and are quoted as published.

Download the data

The aggregates behind every figure on this page are published as a long-format file with one row per statistic: the section, the metric, the group size and, for comparisons, the size and value of the other group. It holds no domain names, hashes, user identifiers or audit identifiers.

• aggregates.csv (CSV)
• aggregates.json (the same rows, with a meta block)

Licence: CC BY 4.0. Cite as: Juan Camilo Auriti, GeoReady (geoready.dev). "State of GEO: October 2026." Published 1 November 2026. https://geoready.dev/state-of-geo/october-2026/

Changelog

• 1 November 2026, v1.0: first publication. 767 domains, 9–31 October 2026, rubric v2.

A correction is added to this list with its date and reason, and the modified date of the page is updated.

Frequently asked questions