How are the scores distributed?
Per this audit, most pages cluster in the seventies — competent but not quotable-on-sight. The tail matters more than the average. Per the same run, a handful of pages score below 60, and those are the ones an answer engine has least reason to repeat.
| Band | Pages | Share |
|---|---|---|
| Easy to quote | 15 | 12% |
| Mostly quotable | 82 | 65% |
| Partly quotable | 28 | 22% |
| Hard to quote | 1 | 1% |
Which check costs the most points?
A pass rate hides cost. A check failed by few pages but worth 18 points can lose more in total than one warned by many pages and worth 4. Per this audit, weighted points lost across all 126 pages ranks as follows.
| Check | Points lost | Share of all loss |
|---|---|---|
| Unsourced numeric claims | 1665 | 46% |
| Direct answer near the top | 413 | 11% |
| External citations | 360 | 10% |
| Sourced figures | 267 | 7% |
| Methodology section | 260 | 7% |
| Publication or update date | 216 | 6% |
| Named author | 184 | 5% |
| Definitions | 164 | 5% |
| Heading structure | 84 | 2% |
| Entity clarity | 24 | 1% |
| Answer-shaped paragraphs | 0 | 0% |
| Tables | 0 | 0% |
Per Gadex's audit, unsourced numeric claims alone accounts for 46 per cent of every point lost across the site. It is also the cheapest to fix, because the figures are already written — they simply have no source beside them.
Which weaknesses travel together?
This is the finding a per-check pass rate cannot show, and the reason the raw corpus was worth measuring rather than sampling. Counting pages that are weak on two checks at once:
| Both weak | Pages |
|---|---|
| Sourced figures + Unsourced numeric claims | 88 |
| Methodology section + Unsourced numeric claims | 70 |
| External citations + Sourced figures | 57 |
| External citations + Unsourced numeric claims | 56 |
| Publication or update date + Unsourced numeric claims | 50 |
| Methodology section + Sourced figures | 48 |
| Direct answer near the top + Unsourced numeric claims | 44 |
| Publication or update date + Sourced figures | 33 |
Per this audit, the top pair is not two problems. Unsourced numeric claims and sourced figures are the negative and positive halves of one behaviour: writing a number without saying where it came from. 88 of 126 pages do it. One editorial habit — attach the source, or drop the figure — moves both checks and, per the table above, most of the recoverable points on the site.
The caution that belongs with it: co-occurrence is not causation, and a single-site corpus cannot separate a general pattern from one company's writing habits. Treat it as a hypothesis to test against your own pages.
Which of our own pages score worst?
Named rather than aggregated. A study of your own site that finds your own site excellent is marketing with a chart.
| Page | Score |
|---|---|
| /reports/sample-gap-report/ | 30 |
| /blog/hourly-freelancers-cannot-build-the-community-trust-you-need/ | 48 |
| /tools/competitor-search-gap-report/ | 49 |
| /blog/content-writer-for-saas/ | 51 |
| /blog/in-house-seo-writing-costs-more-than-you-think-in-the-long-run/ | 51 |
The homepage is on that list. It has no byline, no visible date and no methodology section, and it answers its question in the second paragraph rather than the first — all of which the scorer reports and none of which is hard to fix. It is on the list because the audit runs on every page, not on a flattering selection.
How we measured this
Every figure on this page is generated by a script in this site's repository, which runs the scorer published at /tools/ai-search-visibility-checker/ over the built HTML of the site. It reads each page's main content region only — nav, header and footer are excluded, because a visitor pastes an article rather than a whole document. Per the Gadex audit, legal boilerplate, the 404 page and generated tool-result renderings are excluded as pages where a score would be meaningless, which leaves 126 pages.
Nothing here is typed by hand, and the build fails if the published figures stop matching what the site scores. Scores are computed as weighted marks earned over weight available, so an unscored check cannot cap the reachable maximum. The twelve checks and their weights are documented individually at /checks/.
Limits. One site, 126 pages, one subject area, one set of templates — not a benchmark. The scorer is pattern matching rather than comprehension: it cannot tell whether a sourced claim is correctly sourced, so a page can reach a perfect score while being wrong. Co-occurrence figures describe association, not cause. And none of it predicts whether an answer engine will cite anything.
Why these properties matter for retrieval draws on the platforms' own published guidance rather than on Gadex's opinion:
Common questions
What exactly was measured?
The citation-readiness scorer published at /tools/ai-search-visibility-checker/ was run over the built HTML of all 126 content pages on gadex.ai, reading each page's main content region only. Per the Gadex audit, the mean score is 71 out of 100 and the median is 71, across a range of 30 to 91.
How big is the sample, honestly?
126 pages on a single site. That is small and unrepresentative — one company's writing habits, one template, one subject area. Treat the co-occurrence pattern as a hypothesis worth testing on your own corpus, not as an industry benchmark.
Why publish your own worst pages?
Because a study of your own site that finds your own site excellent is marketing with a chart. The weakest page here scores 30 and the homepage scores 55. Those numbers are more useful to a reader than the average, and withholding them would make the rest untrustworthy.
Can we reproduce this on our own site?
Yes, one page at a time, with the same free tool — no sign-up and nothing is transmitted. The bulk measurement is a script in this site's repository rather than a product; the per-page result is identical either way.
Do these scores predict AI citations?
No. Citation depends on the model, the prompt, the index, the competing sources and a domain's standing, none of which is visible in markup. This measures how easy a page is to quote accurately, which is a precondition rather than a prediction.
Where do you start?
- Run the checker on your most valuable page. It runs in the browser and transmits nothing.
- Fix unsourced numeric claims first — heaviest check, and usually satisfied by attaching a source to a figure already on the page.
- Work down the weight order at /checks/ rather than by which fix looks easiest.