← Back to blog
7 October 2026Kody

Is AI-written content hurting your search rankings?

Google, LinkedIn, Meta and YouTube are all pulling back on content that could have come from anyone. Here is what the data shows, and how to keep using AI without losing rankings or reach.

We talk about AI a lot, so a post warning you about it might seem odd coming from us. But we use it every day, and that's exactly why we want to tell you where it's starting to cost people.

For a while, the playbook was simple. Feed your brand guidelines into an AI tool, get something that sounded like you, and let it produce most of your content. It worked, up to a point. That point has now passed.

Google, LinkedIn, Meta and YouTube are all pulling back on AI-written content, mainly when it reads like it could have come from anyone. Over the last six months we've watched sites that lean heavily on AI lose search visibility, some of them sharply. SEO specialists and the platforms themselves have been talking about it on LinkedIn all year, and the sources are easy to find. We've listed ours at the bottom.

What Google is saying

It started with quality

The first complaints were about quality. Plenty of people were generating whole articles from a one-line prompt, and the results were padded and forgettable. Google's position is that it rewards helpful content whatever tool produced it, but a quick AI draft rarely gives a reader anything the other pages on the topic don't already cover.

Google's guidance on helpful content asks whether your content "provide[s] original information, reporting, research, or analysis" and whether it "clearly demonstrate[s] first-hand expertise and depth of knowledge." An AI working from a short prompt can only draw on what has already been published, so it struggles on both.

Then it went into the rater guidelines

In 2025, Google updated the guidelines for its quality raters, the people who check that its ranking systems are surfacing the right pages. Raters are now told to give the lowest possible rating to pages where the main content is "auto or AI generated… with little to no effort, little to no originality", and to mark down filler.

Those scores are used to test the ranking systems, which tells you what those systems are being trained to recognise.

What the data shows

In July 2026, Ahrefs analysed around 150,000 pages ranking in Google's top ten across 100,000 searches, and ran each one through an AI detector.

2–3×

more impressions for pages with low or moderate AI content than for mostly-AI pages.

40% vs 49%

of mostly-AI pages were indexed, against low-AI pages.

5.3%

of pages in the top three positions were fully AI-generated, so it can still rank.

The researchers put the gap down to quality: "AI content and bad content overlap a significant amount of the time."

An earlier study by Neil Patel's team tracked 744 articles across 68 websites for five months. By month five, the human-written articles were getting 5.44 times more traffic. The AI articles took around 16 minutes each to produce against 69 for a human writer, and the time saved didn't make up for the traffic lost.

Refreshing old pages

This one catches people out. Pasting an older service page into an AI tool and asking for a fresher version feels harmless. The result usually reads more smoothly, but it also tends to drop the client example or the particular way you do things and replace them with general statements that could sit on any competitor's site. Those specifics are often what made the page rank.

Google's warning signs for a refresh

  • Changing the date on a page "to make them seem fresh when the content has not substantially changed".
  • "Mainly summarizing what others have to say without adding much value."

A good refresh adds something new, like a recent project or an answer to a question customers keep asking. AI can help you structure it, but the material has to come from you.

What's happening on social media

LinkedIn

In May 2026, LinkedIn announced that it now limits the reach of posts that look AI-generated and lack a clear point of view. Laura Lorenzetti, LinkedIn's VP and executive editor, put it this way:

When content appears to be generated by AI and lacks clear perspective, it is less likely to be widely distributed beyond a person's immediate network.

Comments are covered too. LinkedIn is targeting comments posted with little human involvement, and replies that just restate the original post. It says early testing identifies generic content with 94% accuracy. AI assistance is still welcome, as long as the voice is yours: "It's ok to use AI to help you write, but your posts and comments need to represent your voice."

LinkedIn had good reason to act. A Pangram Labs study of one million social posts found that more than 40% of long-form LinkedIn posts were fully AI-generated, and LinkedIn accounted for 62% of all the AI content flagged across the platforms studied.

Meta and YouTube

Both are moving the same way.

Meta

In July 2025, it began cutting reach and monetisation for Facebook accounts that repeatedly post unoriginal content.

YouTube

It tightened its monetisation rules for mass-produced videos around the same time.

Instagram

In January 2026, Adam Mosseri said the platform will work to "verify authentic content, and highlight original creators."

What to take from it

You don't need to be posting in bulk to be affected. If a single post looks and feels like everyone else's AI-assisted post, it can hit the same wall.

You may hear that platforms boost content made with their own built-in AI tools. No platform has confirmed that and we haven't found evidence for it, so we'd treat it with caution. The confirmed trend is that platforms favour original content with a clear point of view.

Your AI drafts may be watermarked

How it works

Text from the major AI tools can now carry an invisible watermark. It's built into the words themselves: when the tool picks between words that would work equally well, it uses a secret key to make those choices follow a pattern. Readers see normal text, and anyone holding the key can test for the pattern.

As OpenAI explains, "it lives in the words themselves, it travels with the text when it's copied and pasted", which includes pasting it into your website or a LinkedIn post.

Who's doing it

Google

Has marked Gemini text with its SynthID system since 2024, and its researchers published the method in Nature.

Anthropic

Has used the same method on Claude since August 2026.

OpenAI

Announced on 5 October that it will start watermarking ChatGPT and Codex text for EU users over the coming weeks.

The driver is the EU AI Act, whose transparency rules have required providers to mark AI-generated content since 2 August 2026. If you publish for EU readers, check how those rules apply to you, including the disclosure requirements for text that informs the public.

Editing weakens it

92% → 66%

Detection after swapping just 10% of the words for synonyms, in OpenAI's testing.

90%+

How often a paraphrasing tool removed Google's watermark, according to researchers at ETH Zürich.

The more of you there is in a piece by the time you publish it, the less of a trace the tool leaves behind.

Why a brand voice prompt isn't enough any more

If you've given your AI tool your brand guidelines and tone of voice, it's reasonable to think that covers you. It was a good first step, and it's still worth doing. It just doesn't solve the problems above, for three reasons.

1. Voice covers tone, not substance

Guidelines shape how the tool says something, while what it says still has to come from you. Google asks for "original information, reporting, research, or analysis" and "first-hand expertise", and LinkedIn limits posts that lack "clear perspective". A voice prompt supplies neither. You end up with generic content that happens to use your favourite phrases.

2. Everyone else is doing the same thing

Most businesses in an industry use the same handful of tools with similar guidelines: friendly, professional, clear, customer-focused. When every accountant's or builder's blog reads the same, there's little reason for Google or a reader to favour yours.

This also ties back to which plan you're on. Some lower-tier subscriptions let the platform train on your content, so your original thinking can end up feeding the same systems your competitors use. We cover which platforms do this, and at which tier, in our post on AI tools, their tiers and the NZ Privacy Act.

3. The patterns sit underneath the voice

AI detectors, like the one in the Ahrefs study, look at statistical patterns in how text is built. A tone prompt changes the surface wording and leaves much of that pattern intact. Watermarks are added at the point of generation whatever you ask for, so a draft in your brand voice is still marked.

Readers also pick up the usual AI tells: rule-of-three lists, "it's not just X, it's Y" phrasing, heavy use of dashes and openers that announce what the post is about to do. Most brand guidelines never mention these, so the tool keeps using them.

The test that matters

Would a reader finish your post and think they're glad they found it? If you can't point to the perspective or the value you've added, the AI isn't the only problem.

How to use AI for content without losing rankings or reach

None of this means you should stop using AI. The businesses that keep performing use it to speed up the work around their expertise and keep the expertise itself in their own hands. Here's how we'd go about it.

Before you write

  1. Start with your own material. Before you open an AI tool, jot down the client story, the result, the question a customer asked last week or the opinion you hold. Give the AI those notes. A one-line prompt produces the same post everyone else gets.
  2. Pick topics you can speak to. Google warns against chasing trending topics where you lack real expertise. Write about what you do every day.
  3. Give the AI a list of tells to avoid. Add the phrases and patterns you never want to see. Our guide to the AI tells is a good starting point. It won't catch everything, but it cuts down the editing.
  4. Decide who it's for. Name the reader and the question they need answered. A post written for a specific customer is harder for AI to make generic.

Before you publish

  1. Check every fact. AI tools state wrong figures confidently. Verify the numbers and link to the original source.
  2. Put a real person's name on it. Add a byline and a short bio showing why that person knows the subject. Google's guidance asks whether authorship is clear to visitors.
  3. Be open about how AI was used where it helps the reader, and check whether the EU disclosure rules apply if you publish there.
  4. Keep a simple record. Note which pieces used AI and who reviewed them. It takes a minute and answers the question if a client or regulator ever asks.

While you edit

  1. Treat the output as a first draft. Rewrite it in your own words. This is where most of the value gets added, and heavy editing leaves much less of an AI fingerprint.
  2. Add first-hand detail. A specific example, a real number from your own work, a photo you took, or a mistake you've seen clients make. AI can't invent these and competitors can't copy them.
  3. Take a position. Say what you'd recommend and why. A clear point of view is what LinkedIn says it wants, and it gives readers a reason to remember you.
  4. Cut the filler. Drop introductions that restate the headline and conclusions that repeat the post. Google has no preferred word count, so write until the point is made.
  5. Read it aloud. If it doesn't sound like something you'd say to a client across the table, keep editing.

When you update your website

  1. Add before you reword. Start with what's new, like a recent project or changed pricing, then use AI to help fit it in.
  2. Check Search Console first. See which pages and phrases already bring in traffic, and protect them.
  3. Only change the date when the content has substantially changed.
  4. Review bulk pages. Improve the thin ones or merge them into fewer, stronger pages.

On social media

  1. Write the opinion yourself. Let AI tidy the wording, but the view and the story behind it should be yours.
  2. Reply to comments yourself. A short, genuine reply does more than a polished generic one, and automated or restating replies are exactly what LinkedIn is targeting.
  3. Post less if it means posting better. One post with a real insight will outperform several generic ones that get held back.
  4. Show the work behind the post. A photo from a recent job or a lesson from a client project gives people something no AI tool could have written.

After any Google update or big content change, check Search Console and your LinkedIn analytics. A drop on specific pages or posts usually shows you which content needs attention first.

The bottom line

We still think AI is a very useful writing tool. We're just more careful about what we hand it and what we keep for ourselves.

If you'd like us to review your blogs, website pages or social content for these risks, get in touch and we'll set up a time.

Sources

Point of view

Tell us what’s slowing you down.

A short conversation, then a clear plan for what we’d build. No pitch, no jargon.

Show us what you’re working with →Book a 30-min call