Does AI content rank? Yes, and it does so every day, on terms worth money, on sites you have heard of. The useful question is not whether Google tolerates AI, because it plainly does. The useful question is why so many AI-drafted pages still sit on page four, and what separates them from the ones that do not.

Two sets of numbers get thrown at this question, and they appear to contradict each other. They do not. They measure different things, and once you see that, the fix becomes obvious.

The short answer, and why the two headline findings disagree

Google's published position treats AI as a production method, not a ranking input. Its guidance on AI-generated content says automation earns no special boost and no special penalty: the test is whether the content is helpful, original and made for people. Same test as anything else. Nobody at Google is checking a box marked "written by a machine".

Then there are the detector studies. Detector vendors have published analyses through 2024 and 2025 reporting that content their tools classify as purely AI takes the number one position something like 9% of the time, against roughly 80% for content classified as human. A separate large-scale analysis of top-ranking pages found only around 13.5% were classified as purely human-written. Read those two together and the picture is clear: the overwhelming majority of pages at the top of Google are neither purely one nor purely the other. They are AI-assisted. Worth saying plainly, too, that these studies are run by companies selling detection, and detector classifications are probabilistic guesses about style, with real false-positive rates.

Which is the reconciliation. A detector scores how statistically predictable your sentences are. Google scores whether the page answers the query better than the ten pages already there. Those are different measurements of different things. A page can trip every detector and rank at number one because it is the only page on the internet with the actual figures. A page can read as unmistakably human, written by a real person over two days, and fail completely because it repeats what the current top 10 already say.

Authorship is not the variable. Research depth, first-hand specifics and intent match are the variables. We have written about the practical side of this in our publisher's checklist for Google's AI content guidelines, and the conclusion has not moved: Google does not penalise AI content, it penalises badly researched content.

What actually sinks AI-drafted pages in search

Five failure modes account for nearly all of it, and only one of them is about writing quality.

Regurgitation. A draft produced from a model's general knowledge is, by definition, an average of what is already published. It restates the current top 10 in slightly different words. There is no reason for Google to rank an average of ten pages above the ten pages it averaged.

Intent mismatch. You write a 2,000-word explainer. The top 10 is six pricing pages and a calculator. That article will not rank, however well written it is, because the searcher does not want an explainer and Google already knows it.

Thin coverage of what the SERP is answering. People Also Ask boxes and featured snippets are Google telling you, free of charge, which sub-questions matter. Drafts written without looking at them miss half the brief.

Cannibalisation. This is the quiet killer of high-volume AI publishing, and it is far more common than an outright quality problem. Two posts on near-identical angles compete for the same term, split your internal links, and neither ranks. The site looks busy. The traffic does not move.

Pattern tells. Uniform sentence rhythm, hedged claims with no numbers, stock phrasing, the vague filler paragraph that says nothing. Readers bounce, and bounce is not a direct ranking factor but it is a symptom of the thing that is.

The research layer that has to happen before a word is written

Here is the part most guides skip. The ranking failure almost always happens before the writing, not during it.

A brief should be planned against the live Google top 10 for the target term on the day it is written. Not a topic. A gap. What are those ten pages missing, what format have they settled into, what question do they all dodge. Without that check, you are aiming at a subject and hoping.

Then the numbers. Real UK search volume and keyword difficulty, refreshed often enough to be current, because volumes drift and a term that was viable in March can be a different proposition by October. Then keyword-gap analysis against five to ten crawled competitors: the terms they rank for and you do not, filtered down to the ones you could plausibly win.

Then the check nobody does. Crawl your own site first. Every existing page, every previously published article, checked for duplicate titles and overlapping angles. We do this before every brief, and it is dull, and it prevents the single most common form of self-inflicted damage in AI-assisted publishing.

Finally, format match. The types of search intent behind a query, informational, commercial, transactional or navigational, decide the shape of the page as much as the word count does. If a comparison table is winning, write a comparison. If a calculator is taking the clicks, a guide is the wrong tool. Our research and planning process runs all of this before a draft exists, and that sequencing is deliberate: research is where the ranking is won or lost.

Making the writing read like someone who knows the subject

Take a trades example. A plumbing firm publishes "how much does a boiler service cost"and fills it with ranges like "costs vary depending on your boiler". A competitor publishes the same title, quotes a specific price range for their own area, names Gas Safe registration, and explains exactly what pushes a job to the top of that range: a boiler over twelve years old, restricted access, a flue test that fails. One of those pages is going to win, and it has nothing to do with who or what typed it.

First-hand specificity is the whole fix. Named standards. Real price ranges you are willing to stand behind. The objection a customer raised last month. The step in your process that usually goes wrong. A model cannot know any of that, which is precisely why it has value.

Brand voice is the other half, and it is a measurable input rather than a vibe. We read your website like a new hire would, we crawl it politely, and we build a profile from your own existing copy: typical sentence length, the vocabulary you actually use, the claims you are willing to make and the ones you avoid. On a thin site with three hundred words of copy, the honest move is to say so rather than invent a personality. Then we interview the founder and take the detail from them. You can see the mechanics of that on our how we write page.

Before anything reaches you, a sceptical sub-editor pass challenges every claim and removes every unsupported superlative. Then each draft is graded twice: an SEO Score for whether it is structured to compete, and a Content Integrity score for whether a human expert would put their name to it. Integrity breaks the tie. A piece that scores high on SEO and low on integrity is exactly the sort of filler that gets reassessed in the next quality update, and we would rather hold it.

Publishing choices that decide whether the page gets a fair hearing

Cadence is a signal in itself. A site that published four posts last year and forty in the next fortnight looks like a different site, and not in a flattering way. Ramp gradually. Two a week, sustained, beats twenty in a burst followed by silence.

Keep the technical side boring. Article, Breadcrumb and FAQ schema, output once and only once, which means detecting an existing Yoast or similar install first so you are not shipping duplicate markup. Publish cleanly into WordPress, Webflow or Framer, or export clean HTML or Markdown for Squarespace, rather than pasting a draft that arrives with mangled headings and stray formatting.

Two more that get dismissed as cosmetic and are not. Images built from your real palette, logo and font, sized for blog, Open Graph and social, because generic AI art on a B2B article undercuts the credibility of everything above it. And internal links into a topical cluster, so a new article inherits relevance from the pages around it instead of sitting alone. The full list of what runs on each article sets out where each of these sits in the pipeline.

How to tell if your AI content is ranking, and what to do if it is not

Track weekly, as a trend line. Daily rank checking tells you about volatility, not performance, and position six on mobile with a featured snippet and three People Also Ask blocks above it is worth a fraction of position six on a clean SERP. Our piece on what rank tracking numbers will not tell you goes further into that.

On a low-authority UK SME site, a realistic expectation looks like this: indexation within days to a few weeks, first impressions on long-tail variants around week four to six, and meaningful movement on competitive head terms measured across three to six months of consistent publishing. Anyone promising faster on a new domain is guessing.

When it is not working, the diagnosis splits three ways. Indexed with no impressions means nobody is searching that phrasing, or you have aimed at a term far above your authority. Impressions with no clicks is a title and meta description problem, or you are appearing below a snippet that already answered the question. Clicks with no engagement means the page did not deliver what the title promised, and that is the one to take seriously.

Refresh rather than republish. Add the first-hand detail, update the figures, sharpen the format to match what is actually ranking now. And know when to stop: if the top 10 is all national publishers with ten years of links, the term is out of reach this year. Go after the keyword gap instead, where a crawled competitor is ranking on something specific and winnable and you are not on the page at all. If you want a view of where those gaps sit on your own site, the free content marketing snapshot is the quickest way to see it.

So, does AI content rank? It ranks when the research underneath it is real, the intent match is right and the page says something the current top 10 does not. We do not sell "AI content". We sell content good enough that how it was made stops being the interesting question.

Frequently Asked Questions

Does AI content rank as well as human-written content on Google?

It can, and large-scale analyses of top-ranking pages suggest most of them are already AI-assisted in some form rather than purely one or the other. What decides the outcome is research depth, originality and whether the page matches the search intent Google has already chosen for that query. Authorship is not the variable.

Can Google detect AI-generated content?

Google has not said it runs AI detection as a ranking system, and its published guidance treats AI as a production method rather than a signal. Third-party detectors classify writing style probabilistically, with real false-positive rates, so a detector flag tells you your prose is predictable, not that Google has marked your page. Fix the predictability by adding specifics, not by rewording sentences.

Why do studies show human-written pages taking the number one spot more often?

Because those studies measure detector classifications, and detectors score stylistic patterns rather than usefulness. Pages that read as human tend to carry first-hand detail, original figures and a point of view, and that is what earns the position. The correlation is real, the causation runs through research quality.

How long should it take an AI-assisted article to rank?

On a low-authority UK site, expect indexation within days to a few weeks and first long-tail impressions around week four to six. Competitive head terms are a three to six month project at minimum, assuming consistent publishing and internal linking. Treat anything faster as luck rather than a plan.

How many AI-assisted articles can I safely publish a week?

There is no fixed safe number, but the shape of the curve matters more than the volume. Jumping from almost nothing to forty posts in a fortnight looks like a different website, so ramp gradually and keep every piece checked against your existing pages to avoid cannibalisation. Two to four a week, sustained and properly researched, is a pattern most SME sites handle comfortably.