Generative engine optimization is the work of getting your brand named and cited inside AI answers rather than ranked inside a list of blue links. The evidence so far says it is won less by rewriting pages and more by being mentioned, consistently, across the sources these engines already read.
Earlier this year I spent an afternoon doing something I would recommend to any brand head: I opened four AI assistants, typed in the fifteen questions a serious buyer in my category would actually ask, and wrote down every company name that came back. It took under an hour. The same handful of names kept appearing, and the ranking bore almost no relationship to who is actually largest by revenue. Several companies I would have called market leaders never surfaced once.
That is the part worth sitting with. A discovery layer now exists that a lot of well-run brands are simply absent from, and most of them have not checked.
Why does generative engine optimization matter now rather than later?
Because the traffic maths has already changed, and the numbers are not subtle.
Pew Research Center tracked 68,879 Google searches from 900 US adults in March 2025. On visits where an AI summary appeared, users clicked a traditional search result 8% of the time. Where no summary appeared, they clicked 15% of the time — nearly twice as often. Clicks on the sources cited inside the summary itself happened on 1% of visits.
Ahrefs studied 300,000 keywords and found position-one click-through rates fell 34.5% for terms where an AI Overview appeared, comparing March 2024 with March 2025. It also found that 99.2% of the keywords triggering AI Overviews are informational in intent — which is to say, exactly the queries that thought leadership content is written to win.
And the audience is not a Western phenomenon. Sam Altman confirmed in February 2026 that India has 100 million weekly active ChatGPT users, making it OpenAI's second-largest market after the United States.
The traffic that does arrive this way behaves better. Semrush research published in June 2025 put the average AI-search visitor at 4.4 times the value of an average organic search visitor. Adobe's April 2026 analysis of US retail found AI-sourced traffic up 393% year over year in Q1 2026, converting 42% better, staying 48% longer, and viewing 13% more pages than non-AI traffic.
Fewer visitors, better visitors. That is a trade most B2B marketers would take.
What do AI engines actually cite?
Not what most guides assume. Semrush analysed 230,000 prompts and over 100 million citations across ChatGPT, Google AI Mode and Perplexity over thirteen weeks between July and October 2025. Reddit and Wikipedia dominated overall — but the composition moved violently. On ChatGPT specifically, Reddit's citation frequency fell from roughly 60% to around 10% in mid-September, and Wikipedia from about 55% to under 20%, with LinkedIn and Forbes picking up the slack. Google's AI Mode showed a different mix entirely, with LinkedIn at around 15% and Wikipedia appearing in roughly 2% of responses.
Two things follow from that. First, the citation surface is not one surface; it is three or four with different tastes. Second, it is unstable enough that any tactic tuned to last quarter's mix is depreciating while you implement it.
Then there is the finding that reframed this whole thing for me. Ahrefs ran a correlation study across 75,000 brands to see which signals track with being mentioned in AI Overviews. Branded web mentions came out strongest at 0.664. Branded search volume followed at 0.392, Domain Rating at 0.326, and the number of backlinks trailed at 0.218. About 26% of the brands studied had no AI Overview mentions at all. The authors are careful to note these are correlations, not causes, and that most sit in the moderate-to-weak band.
Still, the ordering is instructive. The thing SEO has spent twenty years accumulating — links — correlates least. The thing brand marketing has always been about — being talked about by name — correlates most.
Generative engine optimization is not a new discipline. It is brand building, measured for the first time by a machine that reads.
The Citation Ladder: four rungs from invisible to cited
I use a four-rung model when I plan this work, because it forces an honest answer to the question of which rung you are actually stuck on. Most teams are not stuck where they think.
|
Rung | The question it answers | What it takes | How long
| **1. Resolvable** | Can the engine tell who you are, distinct from everyone with a similar name? | Person and Organisation schema with a stable identifier, one bio used verbatim everywhere, a complete set of linked profiles | Days
| **2. Readable** | Can a passage be lifted from your page and used as an answer? | Answer-first openings, self-contained sections, tables, dated facts, FAQ markup | Weeks
| **3. Referenced** | Are you named on the sources these engines already read? | Trade press, LinkedIn, community forums, podcasts, industry directories, analyst mentions | Months
| **4. Repeated** | Does the mention recur across many sources over time? | Consistent naming, sustained publishing cadence, repeated third-party coverage | Quarters
Rung one is cheap and almost everyone skips it. If an AI system cannot resolve your brand or your executives to a single entity with defined expertise, everything above it leaks. In my own case there are at least three other people online who share my name, in entirely different professions, and until the schema and profiles said otherwise the engines had no reason to prefer one of us.
Rung two is where nearly all published generative engine optimization advice lives, and it is genuinely worth doing. The original GEO paper from Aggarwal and colleagues at Princeton, presented at KDD 2024, showed that black-box content changes — adding citations, quotations and statistics — lifted visibility in generative engine responses by up to 40%, while noting the effect varies by domain. Adobe's readability scoring found retail product pages averaging only 66% machine-readability against a homepage average of 75%, so there is real, unclaimed ground here.
But rung two is a page-level advantage, and page-level advantages are the ones that get arbitraged away first. Everyone reads the same guides.
Rung three is where the difficulty, and therefore the durability, lives. Being named in trade press, on LinkedIn, in a community thread, on a podcast transcript, in an analyst note — those are the branded web mentions that topped the Ahrefs correlation table. They are also the hardest to fake, which is the point.
Rung four is simply rung three sustained long enough to survive a model update. Given how fast the ChatGPT citation mix moved in a single quarter, a brand that appears in one source once is a rounding error. A brand that appears across many sources repeatedly is a fact about the category.
Why is most GEO advice aimed at the wrong rung?
Because rung two is the only one an agency can invoice for in ninety days. Schema, headings, FAQ blocks and answer-first paragraphs are legible, auditable and fast. Getting your company named in eleven credible places over six months is slower, less attributable and harder to sell.
It is the same asymmetry that has always pushed marketing budgets toward the measurable half of the job. I have argued before that B2B brand awareness is the wrong metric to track, and this is that argument arriving from a new direction: the machines are now rewarding mental availability, they are just calling it citation frequency.
There is also a quieter cost to over-indexing on rung two. If every brand in a category publishes structurally identical, answer-first, statistics-dense content, the format stops being a differentiator and the engine falls back on which name it has seen most. Which returns you to rung three.
What this means for you
A practical sequence, in the order I would run it:
- Measure before you optimise. Write down fifteen questions a buyer in your category would ask an assistant. Run them across ChatGPT, Gemini, Perplexity and Claude once a month. Log which brands appear and whether you are one of them. This costs twenty minutes and it is the only baseline you will get, since there is no Search Console for AI citations.
- Fix entity resolution first. Organisation and Person schema, one bio used verbatim across every profile, consistent job titles, and every owned profile linking back to your site. This is a one-week job that most teams have never done.
- Make the top twenty pages extractable. First 40 to 60 words answer the question in the title. Sections that make sense pulled out on their own. One table per page. A dated statistic every few hundred words. FAQ markup.
- Budget for rung three explicitly. Two or three bylined pieces in genuine trade publications a year, a consistent LinkedIn presence from named executives rather than the logo, and honest participation in the communities where your buyers already argue. Treat it as brand spend, because it is — the same conversation as how much a B2B company should spend on brand.
- Hold the cadence. Recency is a filter, not a tiebreaker. Publishing quarterly and expecting citations is like running one flight of advertising and expecting salience.
None of this replaces the rest of the job. The AI layer sits on top of a buying process that still runs through people, and if the customer journey underneath it is broken, being cited more often just delivers better-informed prospects to a worse experience. It is also worth being clear-eyed about what has and has not changed, which is more or less what I argued about AI in brand management more broadly.
Frequently asked questions
What is generative engine optimization?
Generative engine optimization is the practice of increasing how often a brand is named and cited inside AI-generated answers from systems like ChatGPT, Perplexity, Gemini and Google AI Overviews. It combines entity clarity, machine-readable page structure and third-party mentions, rather than the ranking-position focus of traditional search optimisation.
How is GEO different from SEO?
SEO competes for a position in a ranked list of links. GEO competes to be the source an answer is built from, where there is no list and often no click. SEO rewards links and rankings; the available evidence suggests generative engines lean more heavily on branded mentions, entity clarity and extractable passages.
Does generative engine optimization actually work?
The Princeton GEO study presented at KDD 2024 found that content changes such as adding citations, quotations and statistics raised visibility in generative engine responses by up to 40%, with effects varying by domain. That is a page-level gain. Broader visibility appears to track branded mentions across third-party sources more strongly.
How do you measure AI citations without Search Console?
Run a fixed set of buyer questions across the major assistants monthly and log whether your brand appears and whether your site is cited. Check server logs for crawlers such as GPTBot, PerplexityBot and ClaudeBot. Paid tools from Ahrefs and Semrush automate this, but a spreadsheet establishes the trend line.
Should B2B companies block AI crawlers?
For most B2B brands, no. Blocking crawlers removes you from the retrieval layer entirely, which forfeits the citations rather than protecting anything. Publishers with paywalled inventory face a genuine trade-off. A company whose content exists to build category authority does not.
The part I am still unsure about
Whether any of this stays true for more than a year. The citation mix on one platform moved by fifty percentage points in a single quarter, and the correlations we are all quoting are moderate at best. The four rungs feel durable to me because they describe how reputation has always worked, not how one model currently retrieves. But I would rather be told I am wrong now than discover it in 2027.
So a question for anyone running a brand function right now: have you actually run the query set for your own category, and did your company come back? I would like to know how many people find what I found. Tell me on LinkedIn.
