Why Unedited AI Content Is Now a Ranking Liability

The incentive was impossible to ignore. A tool that turns a headline into 1,500 publishable words in ninety seconds changes the economics of content marketing completely, and pretending otherwise would have been commercially negligent. What the 2026 core updates established is that the economics changed for your competitors too, which means output volume stopped being an advantage at almost exactly the moment it became easy.

Why this matters now

Both 2026 core updates pointed the same direction. Original, first-hand, credible content gained ground. Thin, templated or mass-produced pages lost it. Small businesses with genuine expertise, named authors and useful content were among the clearest winners, and sites leaning on unedited AI content saw the clearest losses.

It is worth being precise about the failure, because the popular framing is wrong. Google is not detecting AI text and punishing it. There is no reliable detector, and Google's own guidance has consistently focused on quality and usefulness rather than production method. What actually happens is simpler and harder to game: a page assembled from what other pages already say contains nothing new, and a search engine comparing it against a page containing genuine first-hand information will prefer the latter.

This is why the phrase to fix on is not AI content but undifferentiated content. An article written entirely by a human that summarises the top ten results has exactly the same weakness. AI simply made undifferentiated content cheap enough to produce at a scale that made the weakness impossible to ignore.

The compounding problem is domain-level. Quality is assessed across your whole site, not page by page in isolation. Forty thin pages do not sit harmlessly in an archive. They lower the average against which your genuinely good pages are judged. Businesses that published aggressively through 2025 frequently found that their best pages declined too, and that is the mechanism.

The commercial cost is easy to underestimate because the production cost was so low. A hundred pages at almost no marginal cost feels free. It is not free if it suppresses the ten pages that were actually generating enquiries.

Common mistakes to avoid

The reactions to this problem are mostly overcorrections in one direction or the other.

• Deleting the entire archive. Some of those pages hold links, rankings and residual traffic. Bulk deletion destroys all of it and frequently makes the situation worse before anything improves. Triage individually.
• Swearing off AI entirely. The tool is not the problem, and a blanket ban costs you real productivity. AI is genuinely good at structure, first drafts, summarising your own material and catching gaps. It is bad at knowing anything about your business.
• Running everything through a humaniser tool. Rewriting synthetic text into slightly different synthetic text adds no information. If the page still says nothing specific, changing its rhythm changes nothing that matters.
• Adding a byline and calling it authored. Attaching a name to a page nobody wrote is worse than leaving it anonymous, because it makes an implicit claim about expertise that the content does not support.
• Publishing more to compensate. The instinct after a traffic drop is to increase output. If the existing library is the problem, adding to it accelerates the decline. Fix the floor before adding to the pile.

There is one mistake that is subtler and more expensive than all of these: continuing to measure content by volume. If your reporting counts articles published per month, you will keep optimising for the thing that caused the problem. The number that matters is enquiries attributable to content, and it is entirely possible for that to rise while publishing volume falls by eighty percent.

The test to apply is uncomfortable but fast. Read any page and ask whether a competent competitor with no knowledge of your business could have written the same page. If yes, it is a liability. If it contains your pricing logic, your actual process, a named client outcome, a number from your own data or a position you would defend in a meeting, it is an asset. Our website copy framework covers how to build that specificity in deliberately.

Step-by-step cleanup plan

This is triage, not a rewrite programme. For a library of a hundred pages it takes a focused fortnight.

1. Export everything with its traffic. Every URL, its organic sessions over twelve months, and its conversions. You cannot make sensible decisions about a page whose performance you have not looked at.
2. Sort by sessions and split into quartiles. Your top quartile is probably fine and should be left alone for now. Your bottom quartile is where the damage is. Start there, not with your favourites.
3. Apply the competitor test to every bottom-quartile page. Could a competitor have written this without knowing anything about your business? Mark each page keep, merge or remove. Be strict. The instinct to defend work you paid for is the main thing that derails this step.
4. Merge before you remove. Three thin pages on adjacent topics become one substantive page. Pick the URL with the strongest signals, fold the genuinely useful content into it, and 301 the others. You keep the link equity and produce something worth ranking.
5. Remove only what is genuinely valueless. Pages with no traffic, no links, no conversions and nothing salvageable. Return a 410 rather than a 404 so the removal is unambiguous.
6. Rebuild the keepers with first-hand material. For each page you are keeping, add at least three things only you could supply. Your real process. A number from your own results. A named example. A position you hold. This is the actual work and there is no shortcut through it.
7. Add genuine authorship. A real person, a real role, relevant credentials, and ideally a sentence explaining why they are qualified to write on this subject. Use the person who supplied the first-hand material.
8. Rewrite the introductions. Undifferentiated content announces itself in the first paragraph, which typically restates the title and explains that the topic is important. Replace it with the specific cost of the problem to the reader. If your opening could be deleted with nothing lost, delete it.
9. Set a publishing floor and write it down. No page ships without at least one first-hand fact, a named author, and something a competitor could not have written. A floor you have not written down is a preference, not a standard.
10. Re-measure after four weeks, then repeat on the next quartile. Recrawling and reassessment take time. Judging the result too early will tempt you into abandoning something that was working.

Expect your page count to fall substantially. That is a sign the process is working, not a loss. A library of thirty pages that each contain something genuine outperforms two hundred pages that each contain nothing, and it is considerably cheaper to maintain.

Where AI genuinely helps is inside this process. Use it to draft structure, to summarise your own case studies into a usable form, to find gaps in an argument, or to tighten prose you have already written. What it cannot do is know what happened on your last project, and that is precisely the information that makes a page worth ranking.

Publishing floor checklist

Apply this to every page before it ships and to every page you rebuild.

• The page contains at least three facts that came from inside your business.
• A named author with a real role and relevant experience is credited.
• The opening paragraph states a specific cost or consequence, not the importance of the topic.
• Every statistic is attributed to a named source.
• At least one example is named and concrete rather than hypothetical.
• The page takes a position that some reasonable person would disagree with.
• No paragraph summarises what was just said or previews what is coming.
• No sentence could appear unchanged on a competitor's site.
• No other page on your site covers the same ground.
• You would be content for a prospective client to read this page first.

If a page fails three or more of these, it does not need editing. It needs merging into something else or removing.

How to measure impact

The reporting has to change alongside the content, because the old metrics will tell you this project is failing while it succeeds.

Stop reporting pages published. It measures effort rather than result, and it rewards precisely the behaviour that created the problem. Replace it with enquiries attributable to content.

Expect total sessions to fall before they rise. You are removing and merging pages, so URL-level traffic will drop. Watch sessions to your retained pages instead, and watch conversion rate across the whole library, which should rise immediately because the denominator got better.

Compare rebuilt pages against untouched ones. Keep a control group. If both groups rise, something external happened. If only the rebuilt group rises, your work caused it. This costs nothing and is the only honest way to attribute the outcome.

Track average engagement time on rebuilt pages. First-hand specifics hold attention in a way generic summaries do not. This metric usually moves before rankings do, which makes it a useful early signal that you are on the right track.

Review quarterly, not weekly. Content quality work operates on Google's reassessment cycle. A weekly dashboard for this will generate anxiety and no information.

Key terms in plain English

Undifferentiated content: A page that could have been written by anyone, about any business, by reading other pages. The actual problem, whatever produced it.

First-hand content: Information that exists because you did the work: your process, your numbers, your client outcomes, your opinions.

Domain quality: Google's assessment of your site as a whole. Weak pages lower the bar your strong pages are judged against.

Content pruning: Systematically merging or removing low-value pages to raise the average quality of what remains.

410 Gone: An HTTP status telling search engines a page has been deliberately removed permanently. Cleaner than a 404 for intentional deletions.

Publishing floor: The written minimum standard every page must meet before it ships. Without one, standards drift toward whatever is fastest.

Control group: A set of pages you deliberately leave untouched so you can tell whether your changes caused the result.

Conclusion and next move

AI did not break content marketing. It removed the last remaining advantage of simply producing more than your competitors, and it did so quickly enough that a lot of businesses were still optimising for volume when the ground moved. What is left is the thing that was always the actual product: information that exists because you do the work and other people do not.

Start this week with your three most recent articles. Highlight every sentence containing something only your business could supply. If the page survives with the rest deleted, you already have your standard. If it does not, you have your first rebuild, and you know exactly what it needs. If rankings are your immediate concern, our guide to recovering from the 2026 core updates covers how the wider visibility picture has shifted.