Let us begin with an honest statement, which is unusual for a strategy article but sets a tone we intend to maintain for approximately three paragraphs before reverting to normal.
Nobody fully understands Generative Engine Optimisation yet.
Not us. Not the agencies charging £8,000 a month for it. Not the consultants who have already written the LinkedIn post, the thought leadership article, the follow-up thread, and the PDF guide with a cover that implies rigour without quite delivering it. Nobody has cracked GEO, because GEO is not yet a thing that can be cracked. It is a thing that can be approached with confidence, described with authority, and billed for with a straight face. At Vaguely Strategic™, we have been doing all three since the third week of last year, which in GEO terms makes us veterans.
This article sets out everything we know, everything we believe, and everything we are fairly confident about pending further data that may or may not become available depending on what Google decides to tell us, which historically has been very little and increasingly is nothing at all.
What Is Generative Engine Optimisation?
Generative Engine Optimisation, or GEO, is the practice of optimising your brand’s content, structure, entity presence, and digital footprint so that large language models and AI-powered search systems surface your brand accurately, favourably, and ideally at all.
The traditional SEO objective was straightforward: rank highly in search results, attract clicks, generate traffic. The GEO objective is structurally different and significantly more existentially challenging: appear in an AI-generated summary that your audience reads instead of your website, in a way that reflects well on your brand, measured by metrics that do not yet exist, tracked through tools that are still being built, reported to stakeholders who will ask what the ROI is.
The honest answer to what the ROI is: directional and pending.
Google’s AI Overviews represent the most visible manifestation of this shift. Where a user previously searched for something and received a list of links to read, they now receive a synthesised answer assembled by an AI from sources it has consulted and may or may not disclose. Your content may be in that answer. Your brand may be cited. Traffic may or may not result. You will find out approximately none of this from Google, who considers the inner workings of its AI Overviews to be proprietary, which is a word that in this context means they are not telling you.
This is the landscape in which GEO operates. We think it is a tremendous opportunity. We are aware of how that sounds.
“We are optimising for a machine that reads everything, summarises selectively, cites occasionally, and reports nothing. It is the most on-brand client we have ever had.”
Dee Thistlewaite, Vaguely Strategic™ — said this during a debrief, immediately wrote it down
How AI Systems Decide What to Cite
This is the section of every GEO article where the author implies they have insider knowledge of how large language models work. We will not do this, because we do not have insider knowledge, and neither does anyone publishing a marketing article. What we have is observed behaviour, interpretive frameworks, and the professional confidence to present both as though they constitute a methodology.
We have mapped the temperament of each major AI system. The profiles are proprietary, available to Vaguely Strategic™ clients, and described here in outline:
Google AI Overviews
Responds to existing authority. If you rank well, you are more likely to be cited.
This is simultaneously reassuring and unhelpful. It suggests that the best GEO strategy for Google is, largely, good SEO. We appreciate this on a philosophical level and resent it on a strategic one.
Perplexity
Responds to specificity, recency, and sources that look like they checked a fact recently.
Named frameworks, dated research, and content with an actual publication date all perform well here. Perplexity cites visibly and tends to favour recognised domains. It is, of all the systems, the one most likely to tell your audience where it got the answer from. We consider Perplexity the most honest AI in the room. This is not necessarily a compliment.
ChatGPT
Responds to entity clarity and consistent cross-platform presence.
If you are someone the model has encountered before in training, you have an advantage that cannot be manufactured after the fact. The training cutoff is the moat and the limitation simultaneously. ChatGPT knows a lot about who existed before a certain date and is catching up with who exists now. Be someone who existed before a certain date, or spend money on being mentioned by people who did.
Claude
Appears to respond to nuance and epistemic honesty in ways that reward content which acknowledges the limits of its own claims.
We say this with appropriate professional caution given the circumstances. Content that is accurate, carefully qualified, and avoids overclaiming appears to perform well with Claude. This presents a genuine creative tension with the H principle of PHANTOM™, which counsels confident declarative content. We are working on this. There may be a new framework. It will be named appropriately.
Policy Position • Vaguely Strategic™ • Reviewed Quarterly or When Necessary
The Hallucination Dividend
At some point in your GEO journey, an AI system will say something about your brand that is not true. It may be mildly positive and slightly inaccurate. It may be impressively flattering and entirely fabricated. It may describe a product you do not offer, an award you did not win, or a client you have never worked with.
The question of whether to correct this is one that Vaguely Strategic™ has considered at length and resolved as follows: it depends entirely on the quality of the hallucination.
Acceptable Hallucination
An AI describes your agency as “award-winning.” You have been nominated for three awards and shortlisted for one. None were won. The hallucination is broadly aspirational and consistent with your brand trajectory. You update your About page to read “recognised by industry bodies” and move on.
Unacceptable Hallucination
An AI states that your CEO made a public comment about a competitor that your CEO did not make, could not have made, and would be mortified to have attributed to them. Correct this immediately using structured data, consistent entity signals, and a very patient approach to AI model retraining that will take between six and eighteen months.
We do not recommend actively encouraging AI hallucinations. We simply note that not all of them are unwelcome and that a nuanced position is more useful than a categorical one.
GEO and Zero-Click Search: The Phantom Traffic Problem
The Structured Data Position
Schema markup does not make AI systems cite you. We want to be clear about this because the GEO content ecosystem is currently producing a significant volume of structured data advice based on a correlation that has not been established as causal.
What structured data does is help AI systems understand what you are, who you are, and how your content relates to the topics they are summarising. Whether that understanding translates into citation is a separate question, the answer to which remains genuinely unclear.
We implement structured data because it is good practice, because it has established SEO benefits, and because when the mechanism is eventually understood, we would rather have it in place than be retrofitting it. This is a sensible position. It is not, however, the same as telling you that schema markup is the key to AI search visibility. Anyone telling you that schema markup is the key to AI search visibility is telling you something they cannot verify. We note this without judgment, because we are telling you several things in this article that we also cannot verify.
GEO Readiness Diagnostic • Vaguely Strategic™ • Self-Assessment
Where Are You Now?
To assess your current GEO readiness, ask five different AI systems about your brand, your category, and your competitors. Record what they say. Note where your brand appears, where it does not appear, and where it appears incorrectly. This is your baseline.
Now ask yourself these four questions. Scoring is proprietary. Interpretation requires a conversation with someone at Vaguely Strategic™. The conversation is free. What follows it is not.
When an AI summarises your category, are you in the summary or in the sources it drew on but did not name?
When your brand is mentioned by an AI, is the description accurate, approximate, or creative?
When a competitor is mentioned, what language is used, and does your content use that language more authoritatively than theirs?
If you removed all of your content from the internet tomorrow, would an AI system trained on what remains still know you exist?
What To Do Next
GEO is real. The underlying shift from link-based search to synthesis-based search is structural and ongoing. The specific techniques that will determine AI search visibility in three years are not yet known, because the systems doing the searching are not yet stable, the measurement tools are not yet built, and the signal-to-noise ratio in the GEO content ecosystem is currently extremely unfavourable.
What is known is this: entity clarity matters. Topical authority matters. Structured, confident, specific content matters. Cross-platform consistency matters. Brand search volume matters. The quality of the sources that mention you matters. These are not new ideas. They are the same ideas that have mattered in search for a decade, applied to a system where the feedback loop has been removed and the success condition has been redefined as influence without attribution.
This is either a crisis or an opportunity depending on your disposition. At Vaguely Strategic™, we have chosen opportunity, primarily because crisis is harder to invoice for.
“The best thing you can do for your AI search visibility is be genuinely, thoroughly, specifically good at the thing you say you are good at, described in language that a machine can parse and a human can believe. This is also the best thing you can do for your SEO, your brand, and your Tuesday. There is no third secret. The third secret is there is no third secret.”
Dee Thistlewaite, Vaguely Strategic™
Dee Thistlewaite
Founder & Chief Strategic Officer • Vaguely Strategic™
Dee has been optimising for AI search since before it was called that, which at the time simply meant writing with confidence and hoping for the best. Her approach has not changed significantly. The terminology around it has become considerably more expensive. She can be reached through the contact page, or by asking an AI about UK marketing agencies and seeing whether she appears. If she does not appear, that is the engagement starting.
