We published 541 AI-written posts in 3 months. Here is exactly what Google did with them.

Short answer: yes, AI-written content ranks on Google in 2026, but not where you want it to. Across 541 posts published by an unattended pipeline on four sites between June and August 2026, weekly Google impressions grew from 915 to 8,217. Clicks over the whole period: 43. Average position across the four sites sat between 18 and 41 the entire time. Volume buys impressions. Position buys clicks. Nothing about the writing itself moved the second number.

This is a data report, not a pitch. The pipeline is ours (MarketingSoHigh), the sites are ours (a competitive-intelligence blog, an AI studio, a careers product and this blog), and every number below comes from Google Search Console and our own publishing database. Where a number is an estimate, it says so.

The setup

• Four sites, all under a year old, none with meaningful backlinks. Domain authority is effectively zero.
• One pipeline: keyword selection, drafting, a factual quality gate, image generation and WordPress publishing, four runs a day, no human in the loop.
• Cadence: 97 posts in June, 177 in July, 165 in August. 199 posts in the last 30 days.
• Length: 3,300 to 3,600 words per post in June and July. We cut that to about 2,000 in August.
• Cost: model spend runs at roughly £2 a day for six to seven posts, so about 30p of generation cost per post. That excludes tooling, the keyword-data API (pennies) and any human time, of which there was none.

What Google did with 541 posts

Three things stand out.

1. Impressions compound. Clicks do not.

Total impressions across the tracked period: 30,779. Total clicks: 43. That is a click-through rate of 0.14 percent, which is exactly what you get when your average position is on page four. Google is willing to show a new site’s page to someone, briefly, on page three or four. Almost nobody scrolls there. The impressions curve looks like success; the clicks curve is the truth.

2. About half the pages get seen at all.

In the week of 24 August, 243 distinct pages recorded at least one impression, out of 495 published by then. The other half were indexed and ignored. When we looked at which half got impressions, it was not the better-scored half: our internal SEO and AI-readiness scores averaged 83 to 97 across every month, with no correlation to impressions. A quality score is a property of the page. Position is a property of the site.

3. Cutting length by 40 percent changed nothing.

In August we dropped average length from about 3,500 words to about 2,000. Impressions rose fastest in exactly that month. Length was not the lever, and the extra 1,500 words were costing tokens and reader patience for nothing.

What was actually wrong

When we audited the output in early September, the problems were not in the prose.

• Titles were 72 to 127 characters long. Google shows about 60. Every post reached the results page with its title cut mid-phrase, because the headline was being pushed to WordPress as the SEO title too.
• Keyword demand was made up. One of our keyword sources recorded our own impression count as the “search volume”, then scored every such keyword a flat 90 out of 100. The pipeline was confidently writing for queries nobody types.
• No FAQ schema was reaching Google on one site, because a table-of-contents plugin injected hidden anchors inside every heading and the schema builder matched the raw HTML instead of the text.
• The entity was fragmented. The Organization on the homepage and the publisher on every article had no shared profile links, so to an answer engine they were two different companies.
• Nothing measured the results page itself. We knew a position; we did not know whether the query showed an AI overview, who it cited, or whose featured snippet sat above us.

All five are fixed as of this week: titles are derived to fit 60 characters and back-filled across existing posts, every keyword now carries a measured Google Ads volume or is excluded, the schema bugs are patched, profile links are delivered to every site’s schema from one place, and a weekly check now reads the actual results page for every keyword we publish on. This report will get a follow-up with the before and after.

What this means if you are considering AI content

1. Expect impressions in weeks and clicks in months, if at all, without authority. A new domain with a thousand good posts is still a new domain.
2. Measure clicks and position, never impressions or quality scores. Both of the latter went up and to the right while nothing happened.
3. Spend the saved money on the bottleneck. At 30p a post, generation is not the cost. Links, brand mentions, original data and pages that answer a question people actually ask are the cost, and they are what moves position.
4. Write shorter. Two thousand words performed the same as three and a half thousand. The extra length is for the writer, not the reader.
5. Publish original numbers. This post exists because a pipeline that only rewrites what is already on the web cannot be cited for anything. Data can. Our AI SEO strategy guide covers what to do once the numbers are in.

Frequently Asked Questions

Does AI-generated content rank on Google in 2026?

Yes. Across 541 AI-written posts on four new sites, weekly impressions grew from 915 to 8,217 in eight weeks, which means Google indexed and showed the pages. It ranked them on pages two to four on average, which is why 30,779 impressions produced 43 clicks. Ranking and getting traffic are different questions.

How much does AI content cost per post?

At our volume, about 30p of model spend per post: roughly £2 a day for six to seven posts, including drafting, a factual quality check and an image. Keyword data adds pennies. The number excludes any human editing because none was done, and excludes the authority work that would have made the posts rank.

How many posts does it take to get traffic?

There is no threshold. We published 541 and received 43 clicks, while impressions rose every week. Traffic followed position, and position followed the site’s authority, not the number of posts. Ten posts on a site with links will outperform five hundred on a site without them.

Is longer AI content better for SEO?

Not in our data. We cut average length from about 3,500 words to about 2,000 in August and impressions grew faster than in any earlier month. Length neither helped nor hurt; it only cost tokens.

What matters more for AI content, volume or authority?

Authority. Volume moved impressions, the vanity metric. Authority, meaning links, brand mentions and pages that answer a real question well enough to be cited, is the only thing that moves position, and position is the only thing that moves clicks.