A decent SEO consultant costs somewhere between a few hundred and a few thousand pounds a month. An AI assistant costs a fraction of that, never sleeps, and will happily read your entire sitemap at three in the morning. Tempting? Of course. But it is also where many site owners go wrong, because a tool that sounds certain is not the same as a tool that is right.
This guide shows where AI genuinely earns its place as a consultant, where it needs supervision, and how to point it at the two things that now matter together: ranking in classic search and being quoted inside AI-generated answers.
SEO and GEO: Two Games, One Pitch
SEO is the old discipline. You earn a position in a list of results.
GEO, or generative engine optimisation, is the newer one. Here the aim is to be one of the sources an AI system draws on when it writes a single, synthesised answer. The term comes from a peer-reviewed paper by researchers at Princeton, Georgia Tech, the Allen Institute for AI and IIT Delhi, presented at KDD 2024.
The two overlap more than the jargon suggests. Google states that AI Overviews and AI Mode need no special optimisation and that existing SEO fundamentals still apply. To be shown as a supporting link, a page must be indexed and eligible to appear in search with a snippet. Miss that, and nothing else you do matters.
What the Evidence Says (and Does Not Say)
Before handing anything to a machine, it helps to know what the humans have actually measured.
The GEO study. The Princeton-led team built a benchmark of 10,000 queries and tested nine content tactics against a generative engine. Their headline result: visibility in generated responses can rise by up to 40 per cent. The tactics that helped most were adding citations, quotations from relevant sources and statistics. Keyword stuffing, the old reflex, offered little to no improvement. The authors also found that results differ by subject area, so a tactic that works for a legal site may do nothing for a recipe blog. One secondary write-up of the paper reports that sites ranked fifth in conventional results saw the biggest relative gains. Treat that as encouraging rather than guaranteed.
Google’s own position. Google says AI Overviews and AI Mode may use query fan-out, which means running several related searches across subtopics before composing an answer. In practical terms, your page is not only competing for the phrase the user typed. It is competing for the questions hiding behind it.
The traffic question. Pew Research Center analysed the browsing of 900 US adults in March 2025. Users who saw an AI summary clicked a traditional result link in about 8 per cent of visits, against about 15 per cent when no summary appeared. Clicks on links inside the summary itself were rarer still, at roughly 1 per cent of visits. So being cited is useful for brand and trust, but it is not a straight swap for a click. Plan accordingly.
Where AI Earns Its Fee
Think of the assistant as a sharp junior analyst. Fast, well-read, and tireless. Also prone to confident nonsense. Give it jobs where you can check the output.
1. Triage a technical audit
Export your crawl data and paste in a sample. Ask the assistant to group issues by what they block and then by how many pages they touch. It is very good at pattern spotting across hundreds of rows, and it will often explain a redirect chain more clearly than the crawler’s own documentation does.
What it cannot do is see your live site. Verify every finding in Google Search Console and your crawler before anyone touches a template.
2. Find the questions you have not answered
Give the assistant your page and ask what a reader would ask next. Then ask what a cautious buyer, a beginner and an expert would each push back on. This mimics query fan-out neatly. Each question you cannot answer from the page is a gap, and each gap is a reason an AI system might cite somebody else.
3. Rework pages so they can be lifted cleanly
Generative systems favour passages that stand on their own. Ask the assistant to flag paragraphs that only make sense with the one before them, to suggest a plain, direct opening sentence for each section, and to point out claims that lack a number, a source or a named example. That last check maps directly onto what the GEO research found helpful.
Keep your voice. Keep your facts. Let the tool suggest structure, not substance.
4. Draft the dull technical pieces
Schema markup, redirect maps, robots.txt rules, title tag variations at scale. These are well-defined tasks with checkable outputs, which is exactly where AI shines. One note: Google says no special structured data is needed for its AI features, so use a schema where it helps a specific search feature, and validate it before it goes live.
5. Stress test your own content
Ask the assistant to answer a set of buyer questions without seeing your page, then again with it. If the second answer barely changes, your page is adding nothing the model did not already know. That is a sobering and useful result.
Where It Will Lead You Astray
It invents things. Statistics, studies, and even Google features that do not exist. Never publish a number you have not traced to its source yourself.
It also flatters. Ask whether your page is good, and it will usually say yes. Ask it to find five reasons a rival page deserves to outrank yours, and you will learn more.
Then there is volume. Google’s guidance on generative AI content is plain: AI can help with researching a topic and structuring original work, but producing many pages without adding value for users may breach its scaled content abuse policy. It points reviewers to the Search Quality Rater Guidelines, sections 4.6.5 and 4.6.6, which cover that abuse and content made with little effort or originality. Reporting on a May 2026 changelog entry indicates Google also clarified that its spam policies apply to generative AI responses in Search. The lesson is dull and dependable. Publish fewer, better pages. Do not let a keyword list dictate your site map.
Finally, a word on confidentiality. Do not paste customer data, unpublished figures or client details into a tool without checking your organisation’s rules and the provider’s data terms.
Prompts Worth Keeping
Adapt these to your own sector. Replace the brackets with real material.
“Here is a page and its target query. List the ten follow-up questions a reader is likely to have. Mark which ones the page answers fully, partly or not at all.”
“Read this section and tell me which sentences make a claim without evidence. Suggest what kind of source would support each one. Do not invent sources.”
“Act as a sceptical competitor. Give me five reasons a rival page could outrank this one for [query].”
“Group these crawl errors by root cause, then rank the groups by the number of indexable pages affected.”
“Rewrite this opening paragraph so it answers the question in two sentences, in plain British English, with no filler.”
Notice the pattern. Each prompt asks for analysis you can check, rather than for finished copy you might be tempted to paste straight in.
Let the Right Crawlers In
A surprising number of sites vanish from AI answers because of one stray line in robots.txt.
OpenAI runs separate crawlers for separate jobs. OAI-Search Bot surfaces sites in ChatGPT’s search features. GPTBot relates to training its foundation models. Each setting is independent, so you can allow the first and disallow the second. Sites that opt out of OAI-SearchBot will not appear in ChatGPT search answers, though they can still show as navigational links. OpenAI also notes that changes can take around 24 hours to take effect.
An example of a selective setup:
User-agent: OAI-SearchBot
Allow: /
User-agent: GPTBot
Disallow: /Whether to block training is a business decision, not a technical one. Make it deliberately. And check that a broad rule or firewall setting is not catching these bots by accident.
Measuring Something That Resists Measurement
Google folds appearances in AI Overviews and AI Mode into the overall Search Console Performance report, under the Web search type. That means you cannot neatly separate them from ordinary results. Be honest about that limit.
What you can do:
- Build a fixed list of 30 to 50 questions your customers really ask.
- Run them monthly in the AI tools your audience uses and record whether your brand is cited, mentioned or absent.
- Watch referral traffic from AI platforms in your analytics, remembering that many answers will never produce a click.
- Track branded search volume. If people start asking for you by name, something is working.
Answers vary from one run to the next, so look for trends over months rather than reading too much into a single result.
A 30-Day Plan
Week one: foundations. Confirm your key pages are indexed and snippet-eligible. Check robots.txt and firewall rules for the crawlers you want. Record your baseline question list.
Week two: audit. Use AI to triage crawl and Search Console data. Verify every finding by hand. Pick the five issues that touch the most valuable pages.
Week three: content. Take your ten most important pages through the gap analysis. Add original data, named examples, expert quotes and properly sourced statistics where the evidence supports them.
Week four: review. Re-run your question list. Compare with the baseline. Decide what to repeat, what to drop and what to hand to a human specialist.
Final Thoughts
AI makes a good consultant for the same reason a good junior colleague does: it is quick, it is available, and it improves enormously with clear direction and a careful reviewer. It makes a poor one when left alone.
The research points the same way as Google’s advice. Be easy to find. Be worth citing. Say something specific, back it up, and make it simple to quote. Do that, and you will be better placed in both a list of blue links and a paragraph written by a machine.
References
Aggarwal, P., Murahari, V., Rajpurohit, T., Kalyan, A., Narasimhan, K. and Deshpande, A. (2024) GEO: Generative Engine Optimization. arXiv:2311.09735 (v3). Presented at KDD 2024. Available at: https://arxiv.org/abs/2311.09735v3 (Accessed: 25 October 2026).
Google Search Central (n.d.). AI features and your website. Available at: https://developers.google.com/search/docs/appearance/ai-features (Accessed: 25 October 2026).
Google Search Central (n.d.) Google Search’s guidance on using generative AI content on your website. Available at: https://developers.google.com/search/docs/fundamentals/using-gen-ai-content (Accessed: 25 October 2026).
Pew Research Center (2025) Google users are less likely to click on links when an AI summary appears in the results, 22 July. Available at: https://www.pewresearch.org/short-reads/2025/07/22/google-users-are-less-likely-to-click-on-links-when-an-ai-summary-appears-in-the-results/ (Accessed: 25 October 2026).
