Guides / Guide

Using AI for SEO without writing for a robot

By Michael Shreeves · 10 October 2026 · Web & SEO · 5 min read

Background

There are two ways to use AI for SEO. One is to have it write a hundred pages and hope. The other is to have it do the dull research and checking work, so you've got more time for the bit only you can write.

The first one is in Google's spam policy by name. "Scaled content abuse" covers generating many pages to manipulate rankings without adding value for people, and Google's own example is using generative AI tools to do exactly that. It doesn't matter how the pages were made. It matters that nobody needed them.

The second one works, and it's how this site is run.

Before anyone sells you "AI search optimisation": Google's own guidance on AI Overviews and AI Mode says there are no additional requirements to appear in them. No special files, no special schema. The same basics as ordinary search apply, and the traffic from AI features shows up in Search Console with the rest of your web search traffic.

Setup

What this site does, as a worked example

Nothing clever, which is the point:

  • Every guide has the same bones: Background, Setup, Edge cases. That's for readers first. It also means each page answers one question in full.
  • Structured data on every page, so search engines know what's a guide, who wrote it and when it last changed. It doesn't move rankings; it changes how a result looks. See structured data basics.
  • An llms.txt file. Google says it doesn't need one, and I've no evidence anything ranks better for it. It costs nothing to keep, so it stays, but that's a different claim from "it helps SEO".
  • IndexNow pinged after each deploy, so Bing and the engines that share its index recrawl changed pages within hours. Google doesn't use IndexNow, so Google still gets a sitemap and, for important pages, a manual indexing request.
  • Search Console for measurement, not guesswork. Which pages get seen, for which searches, at what position.

AI is useful for most of that: drafting briefs, checking titles, spotting missing links, reviewing pages against a checklist. It shouldn't decide what the pages say.

Research and clustering

Give it a list of searches you already have (from Search Console, or your own notes) and ask it to sort them. Don't ask it for search volumes. It will make them up.

Here are search queries people used to find my site, one per line.
Group them into clusters where one page could answer every query in the
group. For each cluster give:
- a name
- the queries in it
- what the person searching wants (to learn, to compare, to buy, to find
  a specific site)
- which existing page of mine fits it best, from the list below, or
  "NEW PAGE" if none does
Don't estimate search volumes or difficulty. If a query doesn't fit
anywhere, put it in a group called "UNSURE".

QUERIES:
[PASTE]

MY PAGES (title and URL):
[PASTE]

Then check a few clusters by searching them yourself. If two queries bring up mostly the same results, they belong on one page. If not, they don't, whatever the model said.

For writing the brief itself, the content brief built on search intent has the prompts.

Titles and descriptions

AI is good at giving you ten options. You pick.

Here is a page's title tag, meta description and its first two paragraphs.
The main search it should answer is: [SEARCH].

Give me 5 alternative title tags under 52 characters and 5 meta
descriptions under 155 characters. Rules:
- Say what the page actually covers. No promise the page doesn't keep.
- Plain words. No "ultimate", "complete" or "everything you need".
- Use the search's own wording where it reads naturally.
Mark which one you'd pick and why, in one line.

PAGE:
[PASTE]

Then count the characters yourself. Models are bad at counting.

Paste a new page and a list of your existing pages, and ask where a link would help a reader. Tell it to skip any link nobody would click.

Here's a new page, and a list of my existing pages (title, URL, one-line
description). Suggest up to 5 places in the new page where a link to an
existing page would help a reader. For each, quote the exact sentence,
name the page to link to, and say in one line why a reader would click it.
Don't suggest a link just because a word matches.

Checking what AI search says about you

Search your main topics in Google, and look at what any AI answer says. If it cites you, check it got you right. If it gets a fact about your organisation wrong, the fix is a clearer page on your site saying the right thing, not a complaint form.

Edge Cases & BCP

  • Nobody can promise you an AI Overview citation. Anyone selling that is selling a guess. Write the clearest answer to the question and measure in Search Console.
  • AI gets facts wrong with total confidence. Every number, date, price and quote in an AI-assisted page gets checked against the original source by a person, and the source is linked.
  • Ten AI-written pages on near-identical topics compete with each other. One good page per question beats several thin ones, for readers and for rankings.
  • Pasting into AI is sharing. Search Console exports and drafts are usually fine. Customer data, analytics with personal data in it, or anything under NDA isn't, unless the tool is one your organisation has approved for it.
  • The rules change. Google updates its guidance, and AI search features move monthly. Recheck the official pages before you quote them.

Checked against Google's own pages on 14 October 2026: AI features and your website and spam policies.

← All guides