From SEO to GEO: A Transition Guide for SEO Teams

Summary

Most of what an SEO team already does - technical hygiene, structured data, topical authority, digital PR - transfers directly to generative engine optimisation. What changes is the unit of optimisation: prompts instead of keywords, citation and mention rates instead of rankings, and continuous multi-engine tracking instead of monthly rank checks. This guide maps what transfers, what to unlearn, and gives a 90-day transition plan you can run with your existing team.

What Actually Changes (And What Does Not)

The most common mistake SEO teams make in their first month of GEO work is assuming the discipline is new. It is not. Generative engines are still reading the open web, still weighting authority and freshness, still preferring sources that are easy to parse. If your site was invisible to Googlebot, it will be invisible to GPTBot, ClaudeBot and PerplexityBot too.

What genuinely changes is the shape of the result. A search engine returns ten blue links and lets the user pick. A generative engine returns one synthesised answer, names two to five brands inside it, and cites a handful of sources underneath. There is no page two. There is no long tail of impressions accumulating below the fold. You are either in the answer or you are not, and the gap between position four and position eleven - which used to be a gradient - is now a cliff.

The second real change is volatility. A Google ranking is reasonably stable week to week. An AI answer is regenerated on every query and can shift when the model is updated, when the retrieval layer surfaces a different source, or when a Reddit thread from last week starts outranking your comparison page as a citation. Teams that check AI visibility once a quarter are reading noise, not signal.

The SEO Skills That Transfer Straight Across

Before rebuilding anything, take stock of what already works. In our experience the following transfer with no modification at all:

• Technical crawlability. Robots directives, render-blocking JavaScript, server response times and canonical hygiene affect AI crawlers exactly as they affect search crawlers. Most AI crawlers execute little or no JavaScript, so client-side rendered content is often a bigger liability for GEO than it ever was for SEO.
• Structured data. Organization, Product, FAQPage and HowTo schema give models unambiguous, machine-readable facts about you. Schema does not guarantee a citation, but it removes the ambiguity that causes a model to describe you inaccurately.
• Topical authority. Depth still beats breadth. A site with fifteen genuinely thorough pages on one subject is cited more often than a site with two hundred thin pages across ten subjects.
• Digital PR and link building. The mechanism has changed but the value has not. Links used to pass authority; now the surrounding text is itself training and retrieval material. A paragraph about you on a respected industry site is a fact the model can quote.
• Content operations. Editorial calendars, briefs, subject-matter review and refresh cycles all survive the transition intact.

If your SEO programme is healthy, you are not starting from zero on GEO. You are starting from roughly sixty per cent.

Four Habits Worth Unlearning

The remaining forty per cent is where the transition gets uncomfortable, because it means dropping practices that have been rewarded for a decade.

• Writing for dwell time. Long preambles, suspense and the classic "we will get to that in a moment" structure were built to keep humans on the page. Models do not dwell. They extract. Front-load the answer in the first hundred words and put the nuance underneath.
• Optimising for exact-match keyword density. Retrieval works on meaning, not string matching. Repeating a phrase eight times does nothing; stating a clear, self-contained, attributable fact once does a great deal.
• Treating your own domain as the only surface that matters. Roughly half the sources an engine cites for a commercial query are third-party: review sites, community threads, roundups, comparison articles. You cannot control those pages, but you can influence whether they exist and whether they are accurate.
• Reporting on traffic alone. AI answers frequently resolve the query without a click. A rising mention rate paired with flat sessions is not a failure - it is often the first sign the strategy is working, and you need the vocabulary to explain that before the traffic conversation happens.

From Keyword List to Prompt Universe

The single most useful artefact in a GEO programme is a prompt universe: a maintained set of the questions your buyers actually type into ChatGPT, Gemini, Perplexity, Claude and Copilot. It replaces the keyword list, but it is built differently.

Keywords are short and stripped of context because search engines rewarded brevity. Prompts are long, conversational and loaded with constraints, because users have learned that context produces better answers. "Best CRM" is a keyword. "What is the best CRM for a 12-person B2B agency that needs HubSpot-level automation without the price" is a prompt. They surface completely different brand sets.

A workable first prompt universe has fifty to a hundred and fifty entries spread across four intent bands:

• Category discovery - "what tools do X", "best software for Y". Highest volume, hardest to win, and the one your CEO will check.
• Comparison - "A vs B", "alternatives to A". The band where a well-built comparison page has the most direct influence.
• Problem-led - "how do I solve X". Where you win on documentation and genuinely useful guides rather than on brand strength.
• Brand-specific - "is A any good", "A pricing", "A reviews". Lower volume, but this is where hallucinated or outdated facts about you show up, and where sentiment damage is easiest to spot and fix.

Write prompts as your buyers speak, not as your marketing team speaks. If nobody outside the company uses the category name you invented, it does not belong in the universe.

The Metrics That Replace Rankings and Clicks

This is the part of the transition most likely to stall, because the familiar numbers stop applying and nobody agrees on what replaces them. Four metrics do most of the work:

• Mention rate - the share of tracked prompts where your brand appears at all. This is your equivalent of indexation: the floor everything else sits on.
• Share of voice - your mentions as a percentage of all brand mentions across the same prompt set. This is the number that tells you whether you are winning or simply present.
• Citation rate - how often your own domain appears in the source list, as distinct from being named in the prose. Being mentioned without being cited means the model learned about you somewhere else, which is fragile.
• Sentiment and accuracy - how you are described when you are named. A first-position mention that misstates your pricing is worse than no mention at all.

Track all four per engine, never blended. ChatGPT, Gemini, Perplexity, Claude and Copilot use different retrieval stacks and different source preferences; a blended average hides the one engine where you are absent. And sample repeatedly - a single run of a prompt tells you almost nothing, because the same prompt asked twice can return two different brand shortlists.

A 90-Day Transition Plan for an SEO Team

Sequence matters more than speed. Measurement first, because everything after it is unverifiable otherwise.

• Days 1-15 - Baseline. Build the first prompt universe, run it across all five engines, and record mention rate, share of voice, citations and sentiment. Do not change any content yet. You are establishing the before.
• Days 16-30 - Audit the sources. For every prompt where a competitor wins, list the pages the engine cited. You will typically find the same fifteen to thirty third-party URLs recurring: review sites, listicles, community threads. That list is your real target list, and it is usually more actionable than any keyword gap report.
• Days 31-60 - Fix extractability. Take your twenty highest-intent existing pages and restructure them: direct answer in the first paragraph, question-shaped H2s, one self-contained fact per paragraph, explicit numbers instead of vague claims, and schema where it applies. This is editing, not rewriting, and it is the highest-return work in the whole plan.
• Days 61-90 - Work the third-party layer. Claim and correct your profiles on the review sites that keep getting cited, pitch the roundups you are missing from, and participate honestly in the community threads that rank. Then re-run the baseline and compare like for like.

Ninety days is enough to see movement on comparison and problem-led prompts. Category-discovery prompts move slower - six months is a realistic horizon, because they depend on an accumulation of third-party signal that cannot be rushed.

How to Explain the Shift to Leadership

The framing that works is not "SEO is dying". It is demonstrably not, and the claim costs you credibility with anyone who reads the traffic reports. The framing that works is that discovery has split into two surfaces, and you are currently only measuring one of them.

Bring evidence rather than an argument. Run ten prompts a prospective customer would plausibly ask, screenshot the answers, and show which brands are named. If a competitor appears in eight of ten and you appear in two, the conversation stops being about methodology and starts being about the gap. That single screenshot deck has unlocked more GEO budgets than any framework we have seen.

Then set expectations honestly: AI visibility moves in weeks and months, not days; the leading indicator is mention rate, not sessions; and the work compounds, because every accurate third-party source you create keeps paying out across every model that retrieves it.