In the study sample published by Ahrefs in July 2026, the higher the AI detection score, the lower the proportion of pages showing indexing signals: 49.28% in the lowest-scoring group versus 40.35% in the highest-scoring group. However, this is a correlation, not evidence that AI content directly causes pages to remain unindexed. These figures are also only lower-bound estimates of indexing rates, not actual indexing rates confirmed by Google. About 40% of pages in the highest-scoring group still showed indexing signals, so the study found no clear indication that AI content was categorically excluded from the index.
The comparison concerns detection scores, not verified proportions of AI-written text
The study used Ahrefs’ AI content detector to estimate the proportion of text classified as AI-generated, then grouped the pages accordingly. The proportions of pages in each group showing at least one indexing signal were:
- AI detection score below 20%: 49.28%.
- 20%–50%: 43.38%.
- 50%–80%: 40.72%.
- At least 80%: 40.35%.
Here, the “proportion of AI content” is a tool-generated estimate, not a proportion verified against authors’ records. For example, a score of at least 80% only places a page in the highest-scoring group; it does not confirm that 80% of its text was actually written by AI. The figures decline across successive groups, but the two highest-scoring groups are quite close. This does not mean that each fixed increase in AI-written text causes a fixed decrease in indexing rates.
Roughly 100,000 pages assessed using three proxy signals for indexing
Ahrefs sampled 1 million pages from its crawler database, with one page per domain. About 100,000 of those pages contained enough text for AI detection and formed the sample for this indexing comparison. The study counted a page as indexed if it met any of the following conditions:
- It had at least one identifiable organic search keyword ranking.
- It had at least one Google Search Console impression since January 2026.
- Its full URL was found using Google’s site: search.
Search Engine Journal’s report specifically cautioned that pages showing these signals were almost certainly indexed, but the absence of these signals did not mean a page was unindexed. For example, a page with no detected rankings or impressions might still be in the index. Thus, the 40.35% figure cannot be taken to mean that Google excluded the remaining 59.65%.
The observed gap does not establish a safe threshold
The study shows differences between groups using its particular sampling and assessment methods. It does not prove that lowering an AI detection score improves the likelihood of indexing, nor does it establish a “below 20% is safe” threshold. Not all pages in the lowest-scoring group showed indexing signals, and not all pages in the highest-scoring group were excluded.
The sample also included only one page per domain and only pages with enough text for detection. The results should therefore not be used directly to predict a particular website’s indexing rate. The study’s authors suggested that content quality might explain part of the gap, but the available material did not establish this causal relationship. This should be treated as a possible explanation, not an official Google rule.
AI content proportions and indexing rates: perspectives from across the web
Why does the report mention roughly 150,000 pages, while the indexing comparison uses roughly 100,000?
The two figures refer to different samples. Search Engine Journal reported that roughly 150,000 pages came from the top 10 results for 100,000 searches conducted in June 2026, with each page containing at least 350 words. The indexing comparison used a separate sample of roughly 100,000 pages, with one page per domain. Ahrefs explained that the latter sample was obtained by drawing 1 million pages from its crawler database and retaining those with enough text for AI detection. The two samples should not be conflated as part of the same comparison.
Do the 2025 study results contradict this indexing comparison?
Not necessarily: the studies examine different metrics. Search Engine Journal noted that the 2025 version covered 600,000 pages and examined the top 20 search results. The correlation coefficient between AI scores and ranking positions was 0.011, almost zero. The current comparison of roughly 100,000 pages instead examined the proportion of pages showing indexing signals in each detection-score group. A page’s ranking position is not the same as whether it shows indexing signals.
What other information should be provided when citing studies like these?
Besides presenting differences between groups, coverage should, as relevant, explain when and how the data was obtained, the sample size, study limitations, potential biases, and information sources. Any applicable margin of error should also be included. Semrush recommends enabling readers to understand the data and make their own judgments, rather than showing them only a single percentage. This is advice on presenting data, not a search platform’s indexing rule.
References
- Google Search Results Still Rank Heavily AI-Flagged Pageswww.searchenginejournal.com · 2026-07-28T18:06:52+00:00 — Discusses: Why does the report mention roughly 150,000 pages, while the indexing comparison uses roughly 100,000?; Do the 2025 study results contradict this indexing comparison?
- Google Doesn’t Punish AI Content; It Punishes Bad Content (331k Pages Studied)ahrefs.com · 2026-07-27T08:00:13+00:00 — Discusses: Why does the report mention roughly 150,000 pages, while the indexing comparison uses roughly 100,000?
- What Google Trends Teaches Us About When To Trust Datawww.semrush.com · 2019-08-30T07:00:00Z — Discusses: What other information should be provided when citing studies like these?

