AI for SEO: work faster without losing Google

What Google says about AI content

The key rule is simpler than many think: Google judges content by whether it helps people, not by whether a person or an AI wrote it. Google said so explicitly in February 2023. Automation was never banned as such; sports scores and weather forecasts have been automated for years.

Google explicitly lists using generative AI tools to produce many pages without adding value for users as an example of scaled content abuse. So are stitched-together content from other pages and pages full of keywords that make little sense to readers.

Where AI really helps with SEO

AI is strongest in SEO where there’s lots of material to sort: hundreds of queries from Search Console, crawl exports, old pages with overlapping topics. Here it saves hours. It’s weakest where you need your own knowledge, current facts and real experience.

What AI does badly in SEO

• Estimating search volume: language models have no access to real search data and like to invent numbers. Volumes come from Search Console or an SEO tool.
• Current facts: prices, rules and data go stale faster than the model knows. Check every number at the source.
• First-hand experience: “we tested” only goes in if it’s true. That’s exactly what sets your page apart from a thousand others.
• Predicting rankings: no AI knows whether a page will rank. Be wary of tools that promise rankings.
• Generic copy: without direction, AI writes what’s already everywhere. That’s the opposite of what ranks.

My rule of thumb: no hype. AI makes good SEO work faster, but it doesn’t replace the reason someone should read your page in particular.

A workflow for one page

Technical SEO with AI

Speed thresholds: web.dev, Core Web Vitals.

Careful with robots.txt: block AI crawlers wholesale and you may disappear from AI answers too. Google-Extended, for example, only controls whether content is used for Gemini training and grounding, and Google says it has no effect on Google Search. Other vendors split their crawlers in similar ways. More in our GEO guide.

My rule of thumb: don’t block AI bots by reflex. If you want to be cited in AI answers, you have to be readable.

AI Overviews: what changes

Since Google started showing AI summaries above the results, users click through to websites less often for some questions. What you can do about it is, per Google, unspectacular: there are no additional requirements and no special optimisation, no new file and no special schema. The page must be indexed and eligible to show with a snippet.

• Allow crawling, including at your host and CDN.
• Key content as text, not only in images or video.
• Internal links, so Google finds the page.
• Structured data that matches the visible text.
• Measure: Search Console counts clicks from AI Overviews and AI Mode in the Performance report under the “Web” search type; Google doesn’t mention a separate filter for them.

If you don’t want to appear in AI summaries, use nosnippet, data-nosnippet, max-snippet or noindex, the same controls as for regular snippets. That does cost visibility in search overall.

Author, disclosure and trust

Google recommends disclosing how content was made where readers would expect it, and not listing AI as the author. A real author with visible experience supports what Google calls E-E-A-T: experience, expertise, authoritativeness and trustworthiness.

• Byline with a real person and a link to an about page
• A short note on how the page was made and who checked it
• An “as of” date on prices and numbers
• Linked sources at the end
• No invented tests or testimonials

Checklist before publishing

• Does the page answer a real question better than the pages already ranking?
• Every number and factual claim checked at the source, with an “as of” date
• Own experience, examples or data you can’t find elsewhere
• Title around 60 characters max, description around 155 max
• Structured data matches the visible text
• Internal links to and from related pages
• Author, note on how it was made, sources
• Checked on mobile, page speed in the green
• Indexing requested or sitemap updated

Metrics

New pages need time. Judge after four to eight weeks at the earliest, and compare pages of similar age, not new ones with old.

Keyword research with real data

AI is only as good at research as the data you give it. Language models don’t know current search volumes, but they sort real data very well. So always pair a data source with the AI.

A simple routine: export three months from Search Console, add customer questions, let AI cluster by intent and flag gaps. Then you decide which two or three pages make the biggest difference. Intent matters more than volume: a low-volume query with clear buying intent is often worth more than a big informational topic.

SEO across markets and languages

If you want to be found in several countries, keep in mind that language models tend to write a generic standard version of a language that doesn’t fit every market equally well.

• Word choice: spelling and vocabulary differ, for example between US and UK English. Adapt AI copy deliberately for each market.
• Legal terms: rules like Germany’s Unfair Competition Act or imprint requirements apply in Germany; other countries have their own laws. AI tends to mix them up.
• Currency and prices: euros, pounds, dollars, with or without VAT, always with an as-of date.
• Multilingual pages: keep languages cleanly separated and link versions with hreflang instead of mixing languages on one page.
• Local examples: tools, vendors and numbers from the target market where it fits.

AI is good at adapting copy for other markets if you tell it who the text is for. “Adapt this for UK readers” gives a better result than “translate”.

Refreshing old content with AI

Improving existing pages often pays off more than writing new ones. Pages that already get impressions but are losing clicks are the best candidates. AI is especially useful here because it quickly checks old copy against new queries.

1. Find candidates: in Search Console, pages with falling clicks or many impressions and few clicks.
2. Compare: AI checks the page text against current queries and flags what’s missing or outdated.
3. Update facts: check prices, dates and rules at the source and add a new as-of date.
4. Cut: remove sections nobody needs instead of adding more and more.
5. Date it honestly: only change the modified date if the content really changed.

A fresh date without a real change isn’t a helpful trick but a trust risk. Readers and search engines notice when only the year changed.

12 prompts for SEO with AI

Always give the AI real data (Search Console export, page text, competitor excerpts). Without data it guesses.

Here’s an export of my Search Console queries with impressions and clicks: [data]. Group them by search intent, name the matching existing page for each group or flag a gap. Don’t invent search volumes.

Customers often ask me these questions: [list]. These are my pages: [URLs with titles]. Which questions does no page answer well yet? Sort by closeness to my offer.

Here are the outlines of the five top-ranking pages for [query]: [outlines]. What do they all cover, what’s missing everywhere, and where could a page with [my angle/my data] be clearly better?

Write an outline for a page on [topic] for [audience]: H2s and H3s, the questions each section answers, and where my own examples or data belong. No filler sections.

Write a draft from this outline: [outline]. Where facts, numbers or first-hand experience are needed, write [check] or [own example] instead of inventing something.

Write five title tags (max 60 characters) and five meta descriptions (max 155 characters) for this page: [text]. Honest, no superlatives, with the main term [term] near the start.

From this text [text], write six FAQ pairs and the matching FAQPage JSON-LD. Only questions and answers that are actually in the text.

Here’s my page list with titles and short descriptions: [list]. For the page [URL], suggest three to five internal links, with anchor text and the spot where the link genuinely helps the reader.

These pages rank for similar queries: [URLs + queries]. Do they overlap in content? Suggest: merge, differentiate more sharply or noindex one, with reasons.

Here’s my Lighthouse or crawl report: [report]. Explain the five most important issues in plain words and sort them by impact and effort.

List every number, date, price and factual claim in this text, with the source that would need to back it: [text].

This page is [months] old and losing clicks: [text + Search Console data]. What’s outdated, what’s missing compared with current queries, and what should I cut?

The most common mistakes

1. Volume over value. Hundreds of AI pages with nothing of your own are exactly what the scaled content abuse policy targets.
2. Invented search volumes. Language models don’t know real search data.
3. Unchecked facts. One wrong number costs trust with readers and search engines.
4. AI as the author. Google advises against it; a real, visible author builds trust.
5. Blocking AI crawlers wholesale. You may vanish from AI answers.
6. Judging too early. New pages need weeks to settle.