State of GEO: August 2026

Score distribution: the floor keeps rising

The average GEO score across 282 unique domains is 56.4 out of 100 — up 2.3 points from July's 54.1 and 2.8 points from June's 53.6. The median is 59, with the bottom quarter of sites scoring at or below 44 and the top quarter at or above 69. The highest score is 100; the lowest is 7. This is the largest single-month improvement in the benchmark's short history, and it happened while the cohort shrank — a point worth unpacking, which the month-over-month section does below.

The band breakdown shows the improvement is broad, not top-heavy:

The practical reading is unchanged in shape but better in degree: roughly seven in ten audited sites still cannot be described, cited, or recommended by an AI engine with confidence. But the Good band grew from 22.8% to 26.6%, Critical shrank from 14.2% to 13.5%, and the Foundation middle thinned from 61.1% to 57.4% — sites are climbing out of the middle tier, not just arriving into it. Three months in, the trend line on "below Good" is 79.2% → 75.3% → 70.9%. Consistent, if slow.

Category breakdown: llms.txt is the biggest gainer

The GEO rubric scores eight categories, each with a different maximum. The table below shows the August 2026 average for each, the maximum, and the efficiency percentage — how much of the available points the average site captures — alongside the June and July efficiencies for context.

The top of the table is stable. Meta tags lead at 90.0% efficiency, Content quality holds at 78.3%, and Robots and crawler access improved to 77.8% — the average site now explicitly allows 23.8 AI bots. These three categories are hygiene: broadly understood, broadly implemented, and not where the score is won or lost.

The movement is in the middle. LLMs.txt efficiency rose from 36.1% to 43.3% — a 7.2 percentage-point gain, the largest of any category, and a full reversal of July's 2.8-point decline. Schema markup continued its climb from 42.5% to 45.0%, and Brand & entity broke its two-month flatline, rising from 38.0% to 41.0% on the back of the Organization schema surge. Every mid-tier category is now trending up together.

The exception, again, is AI Discovery at 10.0% — an average of 0.6 points out of 6, identical to July and barely above June's 8.3%. It is the only category in the rubric that has not meaningfully moved in three months.

Adoption rates: the machine-readability stack

Beyond category scores, the rubric checks specific technical signals. These adoption rates show where the default behaviour of an audited site sits in August 2026 — and what changed from July.

The story of August is that the two AI-specific signals moved together for the first time. In June and July, schema adoption rose while llms.txt fell — teams brought traditional SEO habits but not AI-native ones. In August, llms.txt (+8.6pp), Organization schema (+9.9pp), and FAQ schema (+4.2pp) all climbed at once. The cohort that showed up in August understood both halves of the job.

Three months in: June → July → August

With three editions in the record, the trend is readable. The table below tracks the headline metrics across all three months.

One caveat governs how to read every row in this table. Only 28 of August's 282 domains were also audited in July. Each edition is largely an independent sample of whoever came to GeoReady that month, so a month-over-month delta blends real behaviour change with cohort composition. July's 360-domain cohort followed a traffic spike and skewed toward newer, less-prepared sites; August's 282 skewed the other way. The direction of travel — up — has held for three months, but the month-to-month magnitude should be read as directional.

Among the 28 domains audited in both months, the average score moved from 56.3 to 57.9 — up 1.6 points, with 9 improving, 7 declining, and 12 unchanged. Individual sites move in both directions month to month; the aggregate creep upward is the signal.

The llms.txt rebound — and the stub problem underneath it

July's report flagged a 4.1-point drop in llms.txt adoption and attributed it to dilution: 72 new domains had entered the cohort without AI-specific signals. August confirms that reading. Adoption snapped back to 62.8% — higher than June's 58.3% — as a smaller, more deliberate cohort replaced July's influx. Full, structured files reached a new high of 30.1%.

The gap between "has an llms.txt" and "has a useful llms.txt" is still the real story. Of the 177 sites with a file, 85 publish a complete version and 92 publish a stub — a single line, or a handful of links with no structure. 52% of llms.txt adopters are shipping a placeholder. A stub is marginally better than nothing, but it gives an AI tool almost no orientation: no description of what the site is, no map of the important pages, no context to answer a question about the brand with specificity.

The correlation data explains why the effort is worth it. Sites with any llms.txt average 65.1/100; sites without average 41.8. That 23-point gap is not all causal — prepared teams do many things at once — but llms.txt is consistently the single strongest dividing line in the dataset, and it has been for three months.

The floor that will not rise: AI Discovery at 10% efficiency

Every other category improved or held in August. AI Discovery did neither in any meaningful way. The average score is 0.6 points out of 6 — a 10.0% capture rate, identical to July. Adoption of any AI discovery endpoint actually slipped from 17.5% to 16.7%. Across three editions, the numbers are 8.3% → 10.0% → 10.0% efficiency, and 16.0% → 17.5% → 16.7% adoption. This category is not trending; it is parked.

AI discovery files — /.well-known/ai.txt, and the structured signals inside a complete llms.txt — describe a site's purpose and its most important content in a form an AI can navigate deliberately. Nothing in a default CMS or hosting setup creates them. They exist only when someone decides to publish them, and in August almost nobody did.

That makes it the clearest arbitrage in the benchmark. Publishing a handful of lines at /.well-known/ai.txt moves a site into the top ~17% of the audited sample for this signal, immediately, with no content work and no backend change.

New in August: crawler blocking ticks up

For the first time in the benchmark's three editions, the share of sites blocking AI crawlers moved up month-over-month. 14.2% of audited sites now disallow at least one major AI bot in robots.txt, up from 9.7% in July, and 10.6% block three or more (up from 6.7%). The average site still explicitly allows 23.8 AI bots, so intentional blocking remains a minority position — but the direction changed.

Two forces likely sit behind it. Some of the increase is deliberate: publishers and content businesses reassessing whether unpaid AI crawling serves them, encouraged by machine-readable licensing frameworks like RSL 1.0 that arrived in late 2025. Some is almost certainly accidental — a blanket Disallow rule, a security plugin's default, or a staging config that shipped to production. The benchmark cannot tell the two apart from robots.txt alone.

The practical takeaway for most teams is unchanged: if you want AI visibility, confirm that GPTBot, ClaudeBot, PerplexityBot, Google-Extended, and GoogleOther are not being blocked by a rule you did not write on purpose. If you are deliberately opting out of unpaid crawling, do it explicitly and know which engines it affects.

By TLD: who is leading AI search readiness in August 2026

.com dominates the cohort at 155 of 282 domains and sits close to the overall average at 54.9. The extensions that outperform it are, once again, the ones associated with technical and AI-adjacent teams.

.io leads among TLDs with real sample size: 10 domains, a 66.0 average, 90% llms.txt adoption, and 80% schema. .io is the developer-tooling extension, and the pattern that showed up for .ai in June now shows up here — teams building for a technical audience treat machine-readability as table stakes.

.it stayed strong: 9 domains at a 64.7 average, 100% schema adoption, and — notably — 88.9% llms.txt adoption, up from 50% in July. The Italian cohort has closed the schema-vs-llms.txt gap that defined it in June, when schema sat at 81% and llms.txt at just 33%.

.ai slipped to mid-pack: 20 domains at 58.3, essentially flat on score, but llms.txt adoption fell from 82.4% to 70.0% as the cohort grew and diversified. The AI-native extension no longer has the commanding llms.txt lead it held in June (90%).

.de is the textbook skills-transfer case: 85.7% schema adoption but only 28.6% llms.txt — strong on the mature practice, weak on the new one. .cn remains the lowest-scoring TLD at 35.8, though on only 4 domains; its 25% schema adoption is the real drag. .app was the surprise laggard at 37.8 across 4 domains, pulled down by low adoption on both signals.

What August 2026 tells us about AI search readiness

The clean version of the story: the prepared cohort is pulling away, and it is doing so on every signal at once. In June and July, teams adopted schema (a mature SEO practice) faster than llms.txt (a new one). August is the first month where the AI-specific signals — llms.txt, Organization schema, structured Q&A — all moved up together. The vocabulary gap between traditional SEO and AI readiness is closing, at least among the sites that choose to measure.

The cross-tab makes it concrete. In August, 53.9% of audited sites have both schema and an llms.txt, up from 45.6% in July. The "schema but no llms.txt" group — the classic skills-transfer gap — shrank from 30.0% of the cohort to 22.3%. Teams that invest in structured data are now, more often than not, also publishing an llms.txt in the same pass.

Two things still hold the average down. AI Discovery has not moved in three months — it is a deliberate-publishing problem, and deliberate publishing is exactly what most teams have not gotten around to. And the stub problem persists: more than half of llms.txt files in the dataset are placeholders. Adoption is a vanity metric if the file is empty.

The correlations keep widening. Schema is worth +28.5 points of average score, llms.txt +23.3, FAQ schema +20.2, answer-first structure +19.9. The signals that separate a citable site from an invisible one are becoming more predictive over time, not less — which means the cost of ignoring them compounds.

The August 2026 quick-win checklist

Based on where audited sites lose the most points in August — and the three-month trend — these are the highest-leverage actions, ordered by expected point recovery per hour of work:

1. Upgrade a stub llms.txt to a real one. This is the biggest change since last month's checklist. Adoption is now 62.8%, but 52% of those files are placeholders. If you already have an llms.txt, the win is no longer "publish one" — it is "make it complete": a site description, sections for your key page groups, and real URLs to docs, product, pricing, and blog. Use the free llms.txt generator as a starting structure, then fill it in properly.
2. Add Organization JSON-LD if you are in the 38% without it. Organization schema jumped to 62.1% adoption in August and is the fastest-moving signal in the dataset. Include name, url, logo, sameAs (your verified social and Wikidata/Crunchbase profiles), and description. Pair it with WebSite schema. Sites with schema average 28.5 points higher.
3. Publish a /.well-known/ai.txt. Still the clearest arbitrage: 83.3% of audited sites have not done it, and the number has not moved in three months. A few lines of file place you in the top ~17% for AI discovery immediately.
4. Check robots.txt for crawler blocks you did not intend. Blocking rose to 14.2% in August, some of it accidental. Confirm GPTBot, ClaudeBot, PerplexityBot, Google-Extended, and GoogleOther are allowed — or, if you are opting out on purpose, that you know exactly which engines the rule covers.
5. Structure your Q&A content — but adjust your expectations. Google retired FAQ rich results for all sites in May 2026, so FAQPage markup no longer earns a SERP feature. It still helps AI answer engines lift a clean question-and-answer pair. Keep real FAQ content on the page, mark genuine user Q&A with QAPage where it fits, and do not add FAQPage expecting a Google result.
6. Lead key pages with a direct answer. 78% of audited sites do this, and they average nearly 20 points higher than those that do not. Open your important pages with a self-contained, quotable statement before the marketing context.
7. Re-audit after each change. The rubric rewards compound improvements — llms.txt plus Organization schema plus an AI discovery file is more than the sum of its parts, because a model can triangulate a consistent entity across all three.

Methodology

This report covers 282 unique domains audited by GeoReady between 1 August and 31 August 2026. When a domain was audited multiple times in the period, only the most recent audit is included — this prevents repeat users from inflating the cohort and ensures each domain appears once at its best-known state.

• Engine version: GEO Optimizer 4.15–4.17.x, the open-source 100-point, 8-category rubric. The August cohort was scored predominantly on 4.16.2. Source available on GitHub.
• Audited sample: Sites that chose to audit with GeoReady — not a random sample of the web. Only 28 of the 282 August domains were also in July's cohort, so each monthly edition is largely an independent sample. The cohort skews toward teams already interested in AI visibility, so these figures likely overestimate readiness across all websites.
• Anonymized: Domain names are stored as salted HMAC-SHA256 hashes in the benchmark dataset. No individual domain is identifiable in the public figures.
• Monitoring audits excluded: Only user-initiated audits (web, API, CLI, tools) are included. Scheduled monitoring re-checks are excluded to avoid inflating the cohort with the same domains audited repeatedly.
• Consistent rubric: Every domain is scored with the same engine and the same weights. Category maxima: Robots 18, LLMs.txt 18, Schema 16, Meta 14, Content 12, Brand & Entity 10, Signals 6, AI Discovery 6.
• Month-over-month comparison: June and July figures are drawn from the same benchmark dataset using the same methodology (288 domains audited 10–30 June; 360 domains audited 1–31 July). The cohorts are independent — a domain audited in more than one month appears once per report at its most recent state for that period.
• Efficiency percentages are the category average divided by the category maximum, computed from one-decimal averages, so they may differ by ±0.1pp from a full-precision calculation.

The underlying benchmark dataset is the benchmark_audit_events table in the GeoReady production database. The public API endpoint (GET /api/public/benchmark) returns aggregated figures only — no per-domain data is ever exposed.

Frequently asked questions