Does llms.txt Improve Your GEO Score? What Three Editions of Data Show

Across three State of GEO editions, sites with an llms.txt file score 20 to 24 points higher, on average, than sites without one. That's the single clearest score gap in the dataset — and also the one most likely to get misread as proof that adding llms.txt raises your score. It isn't proof of that, for reasons this piece explains in detail below. Read the gap first, then read why it's a correlation and not a lever.

The gap, three editions running

Every published State of GEO edition to date has run the same "with vs. without" comparison for llms.txt and, separately, for schema markup:

• June 2026 (288 domains, average score 53.6): sites with llms.txt averaged 66.0; sites without averaged 42.0 — a +24-point gap. Schema: 65.0 with vs. 36.0 without, a +29-point gap.
• July 2026 (360 domains, average score 54.1): sites with llms.txt averaged 63.3; sites without averaged 43.3 — a +20.0-point gap. Schema: 61.0 with vs. 32.8 without, a +28.2-point gap.
• August 2026 (282 domains, average score 56.4): sites with llms.txt averaged 65.1; sites without averaged 41.8 — a +23.3-point gap. Schema: 63.2 with vs. 34.7 without, a +28.5-point gap.

The llms.txt gap has held in the 20-to-24-point range for three straight months, and the schema gap — consistently a few points wider — has held just as steadily. For the rest of each edition's findings, see the August 2026 State of GEO edition and the State of GEO hub for June and July.

Why this is correlation, not causation

This is the part that matters more than the numbers above, and it's worth stating plainly: nothing in this dataset shows that adding llms.txt causes a score to rise. Four separate reasons why:

• The scoring formula itself guarantees part of the gap. llms.txt is one of the eight categories the GEO score is built from — up to 18 of the 100 points come directly from that category (August average: 7.8/18, 43.3% efficiency). A site with a complete llms.txt file is credited extra points in that category by definition, before any other signal is considered. Schema markup works the same way, worth up to 16 points (August average: 7.2/16, 45% efficiency). Some of the "with vs. without" gap isn't an external effect of the file — it's built into how the total is calculated.
• Sites that add llms.txt are systematically the sites doing everything else right, too. llms.txt adoption tracks overall GEO maturity — the same teams that bother to publish a curated llms.txt file also tend to have valid schema, clean meta tags, working robots.txt access, and stronger content. The schema gap (+28 to +29 points) moves almost identically to the llms.txt gap across all three editions precisely because both are markers of the same underlying trait — an AI-visibility-aware team — not because either signal independently causes the other's lift.
• This is an observational snapshot, not a controlled test. Nobody added llms.txt to a randomized set of otherwise-identical sites and measured a before/after change. Each edition audits whichever domains chose to run a GeoReady audit that month and splits them by whether they happen to have the file — there is no controlled or randomized measurement anywhere in this dataset.
• The gap describes GeoReady's self-selected sample, not the web. 282 to 360 domains per edition, all opted into an audit — a population that already skews toward GEO awareness. The size of this specific gap may not hold, or even point the same direction, among sites that have never run an AI-visibility audit at all.

The honest read: a maturity marker, not a lever

The most defensible way to read this data: llms.txt presence functions as a marker of GEO maturity, not a lever that pulls a score up on its own. A site that has taken the time to publish a real llms.txt file is, on average, the kind of site that has taken the time to do the rest of the work too — and the score reflects that broader effort, not the file in isolation.

This is consistent with how llms.txt is treated everywhere else on this site: it is not a ranking factor for Google Search, and no AI vendor has publicly confirmed that publishing one changes how often a site gets cited (see What Is llms.txt? for the full framing). The practical implication follows directly: publishing an llms.txt file by itself, without the underlying schema, content, and crawler-access work, is very unlikely to move a score anywhere close to 20+ points — because the file alone isn't what's producing that gap in the aggregate data. It's one visible symptom of a site that's already doing the work, not the cause of the score.

What this data can't tell you

This dataset can't answer the question people actually want answered: if I add llms.txt to my specific site today, holding everything else constant, how much does my score move? That would require a controlled, longitudinal test — the same sites, measured before and after, with nothing else changing.

The closest thing to that in this dataset is thin: only 28 of August's 282 domains were also audited in July, and that overlapping cohort's average moved by +1.6 points overall (9 improving, 7 declining, 12 unchanged) — nowhere near large enough, and not isolated to llms.txt specifically, to support any claim about a single-signal effect. If you're weighing whether to add llms.txt for your own site, treat it as one part of a broader GEO investment, not a standalone lever — and see the llms.txt placeholder problem if you're checking whether an existing file is even complete enough to matter.

See where your own site lands

Run the free audit to see your own llms.txt and schema signals, not just the aggregate averages above.

Run the free AI SEO audit · Build a real llms.txt