AI content
Do AI SEO agents work?
Sometimes, slowly, and less evenly than the headlines suggest. The best public numbers come from a company that sells one: Outrank reports that the median site using its product had about twice its earlier Google clicks after roughly seven months. Its own study estimates those sites would have reached about 1.5 times on their earlier trend alone, and about a quarter of sites ended with fewer clicks than they started with. We sell one of these products too and have no results of our own to show yet, so this page reports other people's numbers and says where each comes from.
In short
- Median clicks in Outrank's data: +7% in the first month, +27% by the third, about 2.0x by the seventh.
- On their earlier trend alone, the same sites would have reached about 1.5x by month seven, by the study's own estimate.
- The top quarter of sites passed 5.4x; the bottom quarter hovered around 1.1x. 24.4% ended below where they started.
- Ahrefs found no meaningful link between how much AI text a page has and where it ranks.
What an AI SEO agent is
A product that does the routine of SEO without being asked each time: it finds keywords, writes an article, publishes it to your site on a schedule, and often builds links and tracks rankings. Outrank, RankPine, RankPill, RankYak, SEObot and our own product are all versions of it. Some vendors call it an agent and others an autopilot; it is the same thing.
| SEO tool | AI writing assistant | AI SEO agent | |
|---|---|---|---|
| What it does | Shows data: rankings, keywords, links | Writes when you ask it to | Plans, writes and publishes on its own schedule |
| Who decides what happens next | You | You | The product, within limits you set |
| Typical examples | Ahrefs, Semrush, Search Console | ChatGPT, Jasper, Koala | Outrank, RankPine, Content Autopilot |
| What you still do | Everything after reading the data | Choose the topic, check the draft, publish | Check what goes out, add what only you know |
The best public evidence
Outrank published Search Console data from its customers' sites, with data to 25 September 2026. It is the only study of this kind we know of with samples and a method, so it is worth reading closely.
On a curve built from 2,119 sites, the median site's clicks were 7% higher in its first 28 days, 27% higher in days 57 to 84, and about 2.0 times higher in days 197 to 224. Growth is slow for three months and then picks up.
The spread is wide. By the seventh month the top quarter of sites were above 5.4 times their starting clicks, and the bottom quarter hovered around 1.1 times. Of 1,383 sites followed for at least 90 days, 24.4% had fewer clicks in their latest 28 days than in the 28 days before they joined, and 8.2% had lost half or more.
The number that matters most
The study gives an estimate that is easy to miss. Before joining, the median site was already growing, by about 5.4%. If that earlier trend had simply continued, the study says, the sites would have reached about 1.5 times their starting clicks by month seven. It does not show how it calculated that.
So the honest comparison is not 2.0 times against nothing. It is 2.0 times against roughly 1.5 times on the path the sites were already on. The gain over that path is about half the starting clicks, not double them. It is still a real gain.
The study lists its other reasons for caution, which is to its credit. Customers chose to sign up, so they are not typical sites. Sites that left are missing from the later months. And in its own words, there is no control group. What the numbers describe is what happened to these sites, not what the product caused.
Does Google hold AI writing against a page?
Not by any measure published so far. Ahrefs looked at about 600,000 pages, the top 20 results for 100,000 keywords, in July 2025. By its own detector, 86.5% contained some AI-generated text. The correlation between how much of a page was AI-written and its position was 0.011, which Ahrefs calls effectively zero. It also notes that pages ranking first tended to have slightly less AI text, and that detectors deal in probabilities.
That matches Google's stated policy, which is about what a page is for and not how it was typed. The longer version is here.
Where it reliably fails
At scale without substance. Otterly, an AI visibility company, built two sites of 1,000 AI-generated posts each. Google removed both from its index, one within 8 days and one within 12 weeks, and ChatGPT never cited either.
One article a day on a question people search for is a long way from a thousand pages at once. The same failure waits at a smaller scale for any product that publishes whatever it generates: articles that say nothing, on topics nobody searches, with facts nobody checked.
What a product cannot do for you
It cannot know what you know. The detail that makes an article worth citing, such as a price you were quoted, a result you measured or a mistake you made, is not on the web for a model to find. It cannot earn real links on its own. A link exchange gives you links. It does not make anyone want to cite you.
What we do not know
We have no results of our own yet: no dataset of customer sites. When we do, we will publish it the way Outrank did: the median rather than the best case, the share that lost, and the trend the sites were already on.
How to judge one before paying
- Ask for the median and the share that declined. A vendor that shows only its best customers is showing you luck.
- Ask what the sites were doing before. Growth against the earlier trend is the number that means something.
- Read what it publishes. Most of these companies run their product on their own blog. Read three posts in a row. If you can see the template, so can your readers.
- Check a claim. Pick a statistic in one of their articles and follow it to its source. If you cannot, their articles on your site will have the same problem.
- Give it three months, with review switched on. The data says the first quarter shows little. Read what goes out under your name during it.
Questions
- How long before an AI SEO agent shows results?
- In the one public dataset, the median site was 7% up after a month, 27% up after three and about double after seven, against about 1.5 times on its earlier trend. A quarter of sites went down. Plan on a quarter before you judge it.
- Is it safer to have a person write everything?
- A person who knows the subject will write a better article. A person who does not, working from the same search results, will write the same article more slowly. The risk is thin content, whoever types it.
- Can an AI agent replace an SEO agency?
- For publishing articles regularly, largely yes. For technical problems, a site move, or deciding what the business should be known for, no.
- What is the difference between an AI SEO agent and an AI writer?
- An AI writer writes when you ask. An agent decides what to write next from keyword data, writes it, publishes it on a schedule and usually tracks the result.
Sources
Each one was opened and read on 14 November 2026. Figures in the text say what was counted, how many and when.
Content Autopilot team
We make Content Autopilot, an SEO and AI visibility product
Our guides are drafted with AI from the sources listed above, and every figure is checked against its source before it is published. Where we sell something a guide discusses, the guide says so.