AI for SEO writing: opportunity or trap in 2026?
What Google has said since 2023 about AI in SEO writing, the uses that break its spam policies, those that help produce useful SEO content and the limits of AI detectors, for a Swiss SME that writes its own pages.
Contents
What Google changed on 5 March 2024
On 5 March 2024, Google launched a core update to its algorithm, designed to show less content that feels like it was made to attract clicks, along with three new spam policies. One of them targets “scaled content abuse”: generating many pages for the primary purpose of manipulating Search rankings, however they are created (Google Search Central).
On the same day, Google said there was “no longer one signal or system” used to identify helpful content. Since August 2022, the helpful content update had added a site-wide signal to identify unhelpful content (Google Search Central). That work now runs through its core ranking systems, using a variety of signals.
A year earlier, on 8 February 2023, Google had set out its position on AI: its focus is “on the quality of content, rather than how content is produced”. Using automation, AI included, with the primary purpose of manipulating rankings does, however, violate its spam policies (Google Search Central).
Google is therefore targeting content with no value produced at scale, with or without AI. For an SME writing its pages in 2026, everything depends on how AI is used: some uses help SEO, others set it back.
What harms SEO
Practice 1: 100% AI writing without revision
Asking ChatGPT, Claude or Gemini to “write me a 1,500-word SEO article on [keyword]” and publishing the text as it stands.
Google points out that generative models don’t retrieve facts: they predict a likely sequence of words, hence inaccuracies, known as “hallucinations”. It asks for all AI-generated content to be fact-checked by hand before publishing (Google Search Central). The length instruction does not help either: Google says it has no preferred word count (Google Search Central).
Practice 2: mass industrialisation
Publishing dozens of generated articles every month to cover every long-tail keyword.
Back in 2022, the helpful content update asked sites whether they were “using extensive automation to produce content on many topics”. Today, Google writes that using generative AI tools “to generate many pages without adding value for users may violate” its policy on scaled content abuse. Sites that violate its spam policies “may rank lower in results or not appear in results at all”. If the site is demoted, your useful articles rank lower along with it.
Practice 3: SEO articles without real expertise
An AI assistant can sound like an expert on any subject, but it sometimes makes up information or repeats generalities. Without a review by someone with real knowledge of the subject, the errors slip through.
For example, an article on the Swiss nFADP that confuses it with the GDPR, or an article on Stripe Connect that credits it with features that do not exist. A reader from the trade spots it, and you lose credibility.
What Google considers useful
Google writes that generative AI “can be particularly useful when researching a topic, and to add structure to original content”.
Practice 1: AI for research and the outline
Use it to:
- Find angles: “Give me 10 possible angles on [topic]”
- Structure the H2/H3 outline: “Suggest a detailed outline for an article on [topic]”
- Find sources: “What recent research exists on [topic]?”, then open each one
- List the claims to check, which you then check yourself in the sources
You then write the text yourself, working from this outline and this research.
Practice 2: AI for the first draft, in-depth human revision
Method:
- A brief written by a person (objectives, angle, sources)
- A first draft generated by the AI
- An in-depth revision: rewriting the weak parts, adding concrete examples, checking every fact and every figure, adjusting the brand tone
- A final sign-off by the person who puts their name to the text
Google also suggests telling readers how automatically generated content was created, where that information is useful to them.
Practice 3: AI for repetitive tasks
- Translation, with a human review for the nuances
- Meta description variants to test
- Summaries and FAQs drawn from an existing text
- SEO optimisation of existing metadata: meta titles that are too long, image alt text
These tasks save time, provided you review what the AI produces. Google asks for generated metadata (title, meta description, structured data, image alt text) to be checked like the rest of the content, because it may appear in search results.
Practice 4: AI for languages you have not mastered
An SME in French-speaking Switzerland that wants to publish in German for German-speaking Switzerland, with nobody on its team who speaks it fluently, can have its pages translated by an AI, then reviewed by a native speaker. Without that review, the page will go live with wrong turns of phrase.
Detecting content generated by AI
Some tools promise to tell whether a text was generated by AI:
- GPTZero
- Originality.ai
- Copyleaks AI Content Detector
The English Wikipedia guide to signs of AI writing advises against relying on them alone: these tools “perform better than random chance” but have “non-trivial error rates”, and text modifications such as paraphrasing can fool them (Wikipedia). A human text can be classed as AI, and a generated text that has been well revised can pass as human.
Google, for its part, judges the real quality of the content, as it wrote in 2023.
The signals that give AI away
The same Wikipedia guide lists several, including overuse of the rule of three. Among the research it cites, a 2025 preprint found that “expert” annotators correctly identified 90% of AI-generated texts, while participants who used language models little did only slightly better than chance. Content that piles up these signals therefore gets noticed, at least by a reader used to these tools.
In an SEO article, they look like this:
- Repetitive structures: “Firstly… Secondly… In conclusion…” in every article
- Empty phrases: “In today’s fast-changing world…”
- Endless lists with no hierarchy: “10 tips”, “15 mistakes”, “20 tools”, with no real expertise behind them
- Rhythmic triplets: “focus, clarity and precision”
- Strings of em dashes for rhetorical pivots, which the guide lists among historical indicators, from 2022 to September 2026
- A lack of concrete examples: everything stays abstract
- Figures thrown in without a source
- A conclusion that summarises instead of concluding
Good practice for human and AI content
Whether you write alone or with help from an AI, these principles apply to any SEO content:
- Figures with their source: “WordPress is used by 39.4% of all websites, according to W3Techs in November 2026”
- Concrete examples: a real client case, an anecdote, precise data
- Explicit citations for figures and technical facts
- A summary at the top of the article, for readers in a hurry
- Regular updates: the article is dated, and pillar pages are reviewed every year
- An author bio: who is writing, with what experience
Among the things that make content trustworthy, Google cites clear sourcing and background about the author. It calls fabricating creator profiles, with AI-generated headshots or made-up names, a form of deception (Google Search Central).
Before publishing SEO content written with an AI, review it with the three questions Google asks of any page: who created it, how, and why. If nobody puts their name to the text and can check every one of its facts, it is not ready.
See our Content service: writing the copy for your website, in your brand voice, with SEO built in.
See our Swiss SEO service: target keywords and the editorial calendar defined at the scoping stage, then a monthly report that adjusts the strategy.
Work out what SEO could bring you: based on your search volume, your target position and your average order value.
Read our article on SEO writers in Switzerland: to choose a human writer.
Read our article on the editorial calendar: to structure your production.