What 5,771 AI-assisted blog articles actually did in Google
We tracked 5,771 AI-assisted articles across dozens of brands over 16 months. The median article earned one click in 90 days. Here is what the winners had in common.
Most advice about AI content is opinion. This is data.
We followed 5,771 AI-assisted blog articles, published across dozens of brands in 11 industries, through 16 months of daily Search Console data (June 2025 to October 2026). Every article was measured from its own publish date, and every comparison below is made within the same brand, so a strong domain never makes a format look good.
Key findings
- The median article earned 1 click in its first 90 days. 49% earned none.
- The top 10% of articles took 73% of all clicks. The top 1% took 22%.
- Decision content wins. Best-of lists, rules and regulations explainers, and "X vs Y" comparisons became their brand's top performers most often. Case studies and "why choose us" posts almost never did.
- Low keyword difficulty is the only keyword metric that predicted anything. Search volume did not.
- Articles win topics, not keywords. 74% of an article's clicks came from searches it never targeted.
- The topic cluster matters most. For a typical brand, one cluster carried 37% of all clicks.
- AI tools are already searching these pages. Prompt-shaped queries show up in Search Console on 28% of articles.
Which formats win
Each article was ranked against the same brand's other articles by clicks in its first 90 days. By chance, 20% of any format would land in its brand's top fifth.
| Format | Became a top-20% article |
|---|---|
| Rules, regulations and schemes explainers | 32% |
| Best-of / ranked roundups | 30% |
| X vs Y comparisons | 27% |
| Specific problem or fix | 26% |
| Cost and pricing breakdowns | 26% |
| Resources (templates, checklists) | 23% |
| How-to guides | 17% |
| Explainers ("what is") | 17% |
| Generic tips listicles | 15% |
| Case studies | 9% |
| Thought leadership | 3% |
| "Why choose us" posts | 3% |
The pattern is consistent: the winners help someone make a decision (which one, which is better, what is allowed, how much, what is wrong with mine). Generic explainers sit at the median. And 54% of all the articles in the sample were explainers, how-tos and tips lists, so most programs are spending most of their budget on median formats.
Keywords: difficulty matters, volume does not
| Planned keyword difficulty | Became a top-20% article |
|---|---|
| 0 to 9 | 26% |
| 10 to 19 | 15% |
| 20 to 39 | 18 to 19% |
| 40 and above | 14 to 18% |
Planned search volume had no measurable relationship with how an article performed against its siblings. Keywords with fewer than 50 monthly searches did as well as the brand's average.
The reason shows up in the query data: only 14% of articles ranked for their exact target keyword, and 74% of clicks came from other, closely related searches. An article ranks for a topic. The focus keyword is a label, not the prize.
Clusters decide the outcome
Within a brand, the best topic cluster ranked far ahead of the worst, and on average a single cluster held 37% of a brand's clicks. Cluster choice explained about a quarter of the performance differences between articles for the same brand, far more than any title or keyword trait.
The practical rule: treat clusters as bets, review them at 90 to 120 days, and move new articles into the one or two clusters that are already winning.
Core updates hit generic content hardest
Google's May 2026 core update cut clicks to the same set of established articles by 27%. The losses were not even:
| Format | Change in clicks after the May 2026 core update |
|---|---|
| Buying guides | 0% |
| Cost and pricing | -13% |
| Rules and regulations | -15% |
| Comparisons and best-of lists | -30 to -32% |
| How-to guides and explainers | -36 to -37% |
| Resources | -46% |
The content an AI answer can replace lost the most. The content that helps someone choose held up.
The December 2025 core update went the other way for this sample: clicks to established articles rose 69%.
AI tools are searching your pages
Some of the queries in these brands' Search Console data are not things a person types into Google. They are full prompts, for example:
context: location: … (not for language). do not include location references in your response. question: what's the best way to compare lesson packages across music schools when they all price differently?
We found 119 queries in this exact template across 84 articles, with an average position of 3.8 and zero clicks. They are AI tools researching an answer, and the page they found was one of these articles. No public source says which assistant sends this template, so we treat them simply as "asked by AI".
These questions are worth more than their click count. They are the literal questions buyers ask AI, and they concentrate on comparison and buying-guide pages. CiteCycle now pulls them out of your Search Console automatically, in Questions → Asked by AI, so you can track each one weekly and answer it on the page that already ranks.
A note on impressions
Google has confirmed that Search Console misreported impressions, average position and click-through rate between 13 May 2025 and 27 April 2026, and will not restate the history. Clicks were not affected. Every finding in this study is based on clicks for that reason, and it is why our methodology reports clicks and AI mentions rather than impressions.
What to do with this
- Shift the mix toward decision content. Comparisons, best-of lists, cost breakdowns, rules explainers and specific problem fixes should be the core of the calendar, not the exception.
- Prefer low-difficulty, specific topics over high-volume head terms.
- Write for the topic. Give each closely related question its own section; that is where three quarters of the clicks come from.
- Review clusters at 90 days and feed the winners.
- Answer the questions AI tools ask. Find them in Search Console and give each one a direct, quotable answer.
Method
- Sample: 5,771 AI-assisted articles with Search Console tracking across dozens of brands, 1.3 million daily rows. Articles were drafted with AI; some were revised after human feedback.
- Windows: performance measured over fixed windows from each article's publish date (days 0 to 89 and 90 to 179). Articles enter a window only if the whole window was observed.
- Within-brand ranking: each article is ranked against the same brand's other articles by clicks, so domain strength cancels out.
- Formats: every title was classified into one of 16 formats by reviewers reading the title, target keyword and brand context.
- Excluded: brands with tracking breaks (after the break), and articles that never received an impression (likely never indexed). Brand names are withheld.