Start with the fact that settles most of the argument. To be used as a supporting link in Google’s AI Overviews or AI Mode, a page has to be indexed and eligible to appear in Google Search with a snippet. Nothing more. Google states there are “no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary.”
Then add the volume. Ahrefs’ prompt index holds 449 million modelled prompts, and 315 million of them are AI Overviews. ChatGPT, Gemini, Perplexity and Copilot are around 30 million apiece.
So roughly seven in ten AI answers people encounter are served by the search engine you were already optimising for, through a pipeline that uses your ordinary index eligibility as the gate. Generative engine optimisation does not replace SEO. It is a layer that runs on SEO’s rails, and a site that cannot get crawled, indexed or snippeted has no GEO strategy at all, whatever it spends.
That is the boring half. Here is where the two genuinely come apart.
The unit of measurement is different, and it is worse
SEO has rank. Position 4 on Tuesday is position 4 for everyone who searches on Tuesday, give or take personalisation, and it is ordinal, stable and auditable.
GEO has mention share, which is a sample. Ask a model the same question twice and you can get two different brand sets, because the output is non-deterministic. There is no position one. Nothing to screenshot and hand to a client as proof. What you have instead is a fixed set of prompts, run repeatedly, producing a percentage that only means something relative to its own history.
The practical consequences are annoying and worth planning for. Your prompt set becomes the asset, so it cannot be edited without resetting the baseline. Sample size is a budget line: 15 prompts tracked daily produces a confident-looking chart built on nothing, and 50 prompts tracked weekly is better data for less money. Every vendor’s headline metric is its own invention with its own formula, so numbers do not transfer between tools and switching costs you your history. The pricing consequences of all that are in what the LLM SEO tools actually cost.
Nobody has solved attribution either. AI referral tracking tells you a visit came from an assistant, which is the closest thing to ground truth available, and it still does not tell you what a citation is worth.
The pages that win are not the pages that rank
This is the biggest practical divergence, and it has a number attached.
Ahrefs looked at which domains ChatGPT cites across a broad sample of US queries in September 2026. Reddit takes 16.8% of mention share. Wikipedia takes 7.0%. Consumer Reports 3.7%, Forbes 3.1%, Healthline 2.5%, YouTube 2.5%.
Now look further down that list, because it is more interesting than the top. engineerfix.com holds 2.8%. legalclarity.org holds 2.8%. Both sit above YouTube. Neither is a household brand, and neither would win a head-to-head ranking fight with Forbes on a competitive commercial term.
Citation share is not gated by domain authority the way rankings are. That is the most encouraging finding in this field for a small site, and it reframes the work: in SEO you want a link from the page that ranks, in GEO you want to be named on it. A listicle that cites you, a Reddit thread that recommends you, a comparison table with your product in a row — those are the raw material models retrieve. Classic link building and this overlap, but the goal shifts from the href to the mention, and an unlinked mention on a frequently-cited page is worth more than a followed link on a page nothing cites.
Which also means the off-site half of GEO is mostly PR and community work, done by people who are not your SEO. Say that out loud before someone adds it to the SEO retainer and then reports on it quarterly.
The payoff is a different currency
Pew Research tracked 900 US adults across 68,879 Google searches in March 2025. When an AI summary appeared, people clicked a traditional result on 8% of visits, against 15% when no summary appeared. Clicks on the sources cited inside the summary happened on 1% of visits. Sessions ended on 26% of pages with a summary, against 16% without. Separately, Ahrefs found across 300,000 keywords that an AI Overview correlated with a 34.5% lower clickthrough rate for the top-ranking page.
SEO buys visits. GEO buys being named in front of someone who is probably not going to click. Both are worth money. They are not the same money, and the mistake is forecasting one with the other’s model.
If you are briefing a board, the honest framing is that GEO is an awareness channel with a 1% click rate and SEO is an acquisition channel being compressed by about a third at the top of informational SERPs. That compression is the actual argument for doing GEO: not that the clicks moved somewhere else, but that fewer clicks exist, so brand familiarity has to carry more of the work.
How I would split the money
For most businesses under €10m revenue: 85% SEO, 15% GEO, and the GEO fifteen is mostly measurement rather than production. You need a prompt baseline, somebody checking it monthly, and your highest-intent pages rewritten so the answer is liftable in the first 60 words. That rewrite is the one task that pays into both columns at once, which is why it goes first. The mechanics are in what AEO actually is.
Do not move budget out of technical SEO to fund GEO. Index eligibility is the gate on both, so that trade is self-defeating, and a crawl problem that costs you rankings costs you citations in the same motion.
The case for spending more is narrow and specific. If your category is one people research conversationally before buying software, insurance, b2b services, anything with a long comparison phase, push toward 70/30, because your buyers are forming a shortlist inside an assistant and you want to be in it. If you sell on price and immediacy, or you depend on local intent, stay at 90/10 and spend the difference on why your traffic actually moved, which is usually something duller than AI.
What I would cut without hesitation: anything that bills for an AI-specific markup standard, an llms.txt file, or a separate GEO retainer whose deliverables are indistinguishable from on-page SEO. Google has said in writing that no new files or schema are needed. Also cut ChatGPT-only reporting. It is 30 million prompts against AI Overviews’ 315 million, and a dashboard that watches the smaller surface because it is the more famous brand has the proportions backwards.
What the report should contain
Four things, and they fit on one page. Impressions against clicks by query from Search Console, which separates “an answer ate the click” from “we lost the ranking”. Mention and citation rate against a prompt set that has not changed. The domains cited instead of you, which is your target list for the off-site work. And AI referral sessions, labelled as directional, because that is what they are.
Everything else in a GEO dashboard is decoration. The metric that matters is whether the cited-instead-of-you list is getting shorter, and that moves on a timescale of quarters.
If you want that baseline established against a fixed prompt set before you commit to a retainer, our AI visibility audit does exactly that, and a fair number of them end with us telling the client their AI visibility is fine and their internal linking is the problem.
