Three tools scored the same 10 sites: why the scores disagree

If you have scanned one site with two different tools, you have seen scores that do not match. This post shows how large the gap can get, why it happens and how to read a score report, using our own measurement.

The measurement ran between the night of 18 October and the morning of 19 October 2026. On the publication day the measurement is 10 to 11 days old. The sample is 10 sites, one night and one pair of exit addresses. Results show a direction and prove nothing.

What did we do?

We picked 10 sites in English and Turkish across content, e-commerce and software. One of them is our own site. We do not publish site names. We used three tools:

• Tool 1: Ruescan, our own tool. It reads the home page, site files and two sample pages from the sitemap.
• Tool 2: a free open-source GEO scorer. It samples eight pages and tests access with 10 different bot identities.
• Tool 3: a free open-source GEO auditor. It reads only the home page.

We do not name the two open-source tools. The aim is to explain the gap. No paid key was used with any tool and no money was spent.

How far did the scores diverge?

Comparing absolute scores makes no sense because the scales differ, so we compared rankings. For ranking agreement we used the Spearman rank correlation (1 is perfect agreement, 0 is none):

Tools 1 and 3 gave the top three sites in the same order. Tools 1 and 2 diverged a lot. In a sample of 10, the position of one site alone moves the correlation noticeably.

Where do the tools agree?

On measured facts the agreement is high. For whether the home page is open to indexing, the canonical, the title, the meta description and whether AI search bots are blocked in robots.txt, Tool 1 and Tool 3 matched on 10 of 10 sites. Both found the same four sites with no H1.

The difference appears when findings are turned into a score.

Why do they diverge?

1. Scope. Tool 3 reads only the home page. Tool 2 samples eight pages. Ruescan reads the home page plus two sample pages. Freshness dates and author bylines show it: Ruescan marked the freshness date "open" on eight sites because it did not find one on the home page. Tool 2 found a date on inner pages for six of them. Each is right within its own scope. "Not on the home page" and "not on the site" are not the same thing.

2. Scale and weight. Tools weigh things differently. Two tools score the llms.txt file. Ruescan does not score it and only shows it as information. The two tools agreed on whether the file exists on 10 of 10 sites. How much it adds to the score differs. We did not measure the effect of llms.txt on citations in this round.

3. Bot access gates. On one site, two AI search bots were fully blocked in robots.txt. Tool 2 put a ceiling on the score and held the site at 40. Ruescan's equivalent ceiling only applied when all three search bots were blocked, so that site got 71. This is a calibration finding. We have not yet decided the threshold.

4. The same tool moving on its own. We ran Ruescan twice on the same day. One site produced no score in the first run because of a redirect loop, so the comparison covers nine sites. Seven of nine kept the same score. One went from 87 to 86 and another from 78 to 75. Changing the exit path moved scores by at most 2 points, except for the two exceptions (the country selector page and the redirect loop). We did not investigate the cause. So do not read meaning into a 3-point gap.

Ruescan's own mistakes

The comparison showed us errors too. We do not hide them:

• On one site the home page redirected to a country selector. Ruescan's direct exit measured that page and scored the site 30. The same code through a different exit path gave 80. Where the client came from changed the result. We could not confirm the exact cause (address or fingerprint). Tool 2 landed on the same selector page and gave 31.
• The home page of one site returned a 301 to itself. On the direct exit Ruescan stopped with a "too many redirects" error after 5 steps and produced no score, which is the right behaviour compared with inventing one. On another exit path the same site was measured. That is why this site is outside the same-day comparison.
• On one site robots.txt listed eight sitemap lines. Ruescan tried only the first, which was a dead address, and counted the sitemap as missing. That was a false alarm.
• On one site the page printed "last updated just now" on every request. Ruescan counted it as proof of freshness. That was a false pass.

We found errors in the other tools too. Tool 3 flagged decorative images with an empty alt as missing. Tool 2 scored a site that fell onto the selector page using a single page and did not say so in its note. None of these findings answers "which tool is right". Each tool is consistent within its scope, and each is wrong in some cases.

How to read a score report

1. Compare with the same tool under the same conditions over time. Comparing scores across tools misleads.
2. Look at the individual findings under the score. Checking a finding like "robots.txt blocks an AI bot" by hand takes minutes.
3. Find out what the tool looks at. A tool that does not state its scope is hard to interpret.
4. Do not react to a gap of a few points.

We published our own site's scan separately in what is still open on ruesandora.com. How we calculate Ruescan's score is on how Ruescan scores.

Frequently asked questions