Will AI replace SEO? No. It is changing who wins.

Indexing means a search engine has discovered a page and may keep it in its index. It does not mean the page ranks for a useful query. Ranking is the order in which search results appear for a particular search. A page can be indexed and still sit too far down the results to bring meaningful traffic. Improving it may involve making the answer clearer, earning relevant references or fixing a site problem that stops search engines understanding the page.

An AI-generated answer is a third thing. Depending on the product and the way it is being used, a system may retrieve current web pages, draw on indexed material, synthesise an answer from its model or combine approaches. Some answers display source links. Others do not. ChatGPT, Perplexity and Google's AI Overviews do not all work in the same way, and their behaviour changes over time. A citation in one answer is not a permanent ranking position.

This distinction matters when someone offers to 'rank you in AI'. First ask which product, which question and which measurable outcome they mean. A page that gets cited, a brand mentioned without a link and a visit from a search result are three separate observations. None guarantees the others.

A buyer can ask a complete question instead of typing two or three keywords. They may ask for a comparison, request a summary of a complicated topic and then refine the question in the same conversation. The final answer can combine ideas from several places. That is convenient for the buyer, but it makes the route from information to website visit less predictable for the publisher.

Consider a finance director researching software for a distributed team. One question may ask what the categories are; another may ask about implementation risk; a later one may ask which supplier has experience with a particular integration. A broad page repeating 'best software' is less useful than a clear explanation of the actual decision. Good content needs to address the question the person is trying to settle, including the limits of your own offer.

Your pages need to be accessible, understandable and worth showing. Check that search engines can reach important pages, that those pages are not accidentally excluded, and that internal links connect your expertise to the relevant services. Give each page a clear purpose. If five near-identical pages all claim to answer the same question, it is difficult for a visitor to know which one to trust.

A direct answer early on helps a human reader as much as any retrieval system. Follow it with the reasoning, trade-offs and evidence needed to make a decision. Name a source when you use someone else's research. Explain how a number was measured when the method matters. None of these is a proven switch that makes an AI tool cite you. They are sound editorial practices whether or not the interface changes again.

Thin content was a problem before AI answers arrived. A page that rearranges the same definitions already available everywhere gives a reader little reason to stay, share or return. Generating more pages at speed does not solve that weakness. It can make it harder to find the one page on your site that contains genuine experience.

The better test is whether your article can answer a question a sales colleague hears and provide something a generic summary cannot. That might be a careful comparison, a worked process, a documented result or a clear explanation of when your approach is a bad fit. It does not require turning every article into a case study. It does require an identifiable point of view and enough detail for somebody to act on it.

For Hampleton Partners, Refiner built a technology M&A reporting programme. Report downloads grew from 200 to 2,000 per month within 12 months, and the programme earned more than 120,000 backlinks over its historical growth period. Those figures show the value of producing material that serves a defined audience and can be referenced by others. They do not show that an AI assistant cited those reports. The work predates much of the present discussion about AI-generated search answers.

Private Dining Experience gives a different lesson. Focused organic work took the business from 1 to 20 organic leads per month in four months, with a 23% organic search click-to-lead ratio. Those are search and conversion outcomes, not AI-visibility metrics. A page has to do something useful after a visitor arrives. More mentions in an answer, if they never bring a qualified conversation or support a later decision, are a weak business result.

Both examples argue for better underlying material and a clear route to enquiry. Neither supports a claim that particular phrasing, schema or a certain number of backlinks guarantees placement in a generated answer.

Start with a small set of pages that matter to real enquiries. For each one, write down the customer question it should answer. Read the first two paragraphs without looking at the title. Can you tell what the page is about and get a useful first answer? If the introduction delays the point to fit a keyword, rewrite it.

Next check the evidence. Are named examples accurate and permitted to be public? Is a number attached to a period and a definition? Does the page explain the awkward part of the decision, such as cost, implementation or what happens if the approach does not fit? Compare the answer with competing pages. If yours offers the same advice in different words, speak to the people delivering the work before adding another paragraph.

Keep traditional search reporting in place. Search Console can show which queries bring impressions and clicks to your pages, and analytics can show what visitors do after arrival. Review those figures by topic, not just across the whole site. If clicks fall while impressions hold steady, investigate the results page and competitors before assuming an AI answer caused the change. Search Console does not give a universal, definitive measure of how often an AI system used your content.

You can also inspect referrals from AI products in your analytics. This only counts visits that actually arrive with recognisable referral information; it misses people who read an answer and never click, and some referral data will be incomplete. Keep those limits visible when you report the number.

For important buyer questions, run a repeatable manual check across the products your audience might use. Record the exact prompt, product, date, whether your brand was mentioned and whether a link appeared. Repeat the same checks over time. Treat the result as a sample, not a market share figure. Answers vary by prompt, location, account settings and day. Third-party visibility trackers can save time, but their simulated prompts and scores are not a substitute for actual enquiries and visits.

First, verify that your most important service and proof pages are accessible and indexed. Fix broken internal routes, retired pages that should redirect and content that says almost the same thing in several places. Next, choose the questions buyers ask before they are ready to contact you. Give each a page with a direct answer, a meaningful explanation and a sensible connection to the relevant service.

Ask the people who speak to customers what prospects misunderstand. Add their answers, with examples you can substantiate. Set up a simple monthly review of search performance, qualified enquiries and a small sample of AI answers. If an AI product changes how it displays sources next month, you will still have a useful record of what real buyers did.

SEO is not going away because the interface has changed. Its job remains helping the right person find and trust the right answer. The methods for observing that journey are becoming less tidy. Be honest about what you know, improve the pages you control and judge success by commercial relevance, not by a screenshot of one generated answer.

GEO stands for generative engine optimisation. People use it to describe work intended to make a brand or its material appear in AI-generated answers. SEO usually refers to visibility in search results that people can click. The labels describe different things you may observe, but the work overlaps: clear, accessible pages, useful evidence and earned references help people assess a source in either setting. GEO is not a separate guaranteed ranking system, and an AI mention without a link is not the same result as an organic visit.

Start with a real buyer question. Answer it plainly near the top, then give the context needed to judge the answer: who it applies to, the trade-offs, an example you can substantiate and a sensible next step. Check that the page is accessible to crawlers and links to the service or proof behind the advice. Do not add speculative schema or mass-produce question pages just to chase citations. Google's guidance says its AI features rely on core Search systems and require no special AI-only optimisation.

Look at the outcome you can actually verify: relevant search exposure, visits from identifiable sources, enquiries and sales conversations. For Google, review the Search Console reports available to your property and read their definitions before comparing periods; for other assistants, referral data only captures visits with a detectable source. Sample a fixed set of buyer questions for mentions and links, but report those observations separately from traffic and pipeline. A citation screenshot is a clue, not a conversion.