GEO should start with what your organisation knows

Technical GEO matters, but it starts too late if the organisation has nothing distinctive to retrieve. The more interesting question is how editorial teams capture specialised human knowledge, turn it into evidence and make each piece of work improve the next.

I should admit that I have a selfish reason for making this argument. If generative engine optimisation turns out to be mainly a technical discipline — crawlability, schema, entity clarity, page structure, citation monitoring and AI visibility tracking — there are people much better equipped than me to lead it. If it turns out to be partly an editorial and organisational-knowledge problem, however, it happens to play directly to the skills I have spent years developing: interviewing people, working with customer evidence, shaping narratives, building thought leadership and deciding what is actually worth publishing.

That does not make the argument true, but it does not make it false either. It gives me a reason to scrutinise it more carefully, especially because I have already argued that AI search risks optimising the wrong layer. The more I work around GEO, the more I think the important distinction is not between technical and editorial GEO, as though one has to defeat the other. It is between making existing information easier to retrieve and deliberately creating stronger knowledge for people and machines to retrieve in the first place.

The scarce thing is not another optimised page

A lot of GEO work understandably begins with the engine. Teams look at the questions appearing in generative search, the sources being cited, the structure of pages that surface, the clarity of entities and whether important information is available in formats an AI system can understand.

All of that can improve discovery. The problem appears when those activities become the whole strategy, because they assume the organisation already has something distinctive, useful and well evidenced to say. A technically immaculate page cannot manufacture expertise, and a perfectly structured answer is still generic if it draws on exactly the same public information as everybody else.

Generative AI makes that problem sharper. It is becoming exceptionally good at synthesising information that already exists, which means another competent summary of a well-covered subject is easier to produce at the same time as it becomes less scarce. The strategic question therefore moves upstream: what does this organisation know that is specific, current, defensible or difficult for somebody else to reproduce?

That question leads away from content production and towards access. It asks which customers, practitioners, subject-matter experts, product specialists, researchers and operational teams are close enough to the problem to know something that is not yet obvious from the public web.

The underused GEO asset may be a real person who knows something

This is the part of the GEO conversation that has surprised me. I expected more emphasis on acquiring current, specialised, first-hand knowledge from real people, rather than concentrating so heavily on how existing pages should be optimised.

Perhaps the reason is simple: getting useful knowledge from people is hard. The right expert may be busy, the customer may need approval, the interviewer needs enough context to recognise an interesting answer, and the resulting claims may require checking before they can be published. Even after the conversation has happened, somebody still has to distinguish between a genuinely useful observation, a rehearsed talking point and a confident anecdote that does not survive scrutiny.

There are also organisational frictions that do not appear in a technical checklist. Legal review can slow a timely answer, customer permissions can take weeks, experts may be comfortable speaking but not writing, and different teams may own fragments of the evidence without anybody owning the full question. A lot of valuable knowledge therefore remains trapped inside calls, transcripts, webinars, events and internal conversations because turning it into dependable public knowledge is labour-intensive.

But that difficulty is also what makes it interesting. Generic information is cheap to reproduce, and technical optimisation can be copied once the practices become widely understood. Specialised first-party knowledge is harder to reproduce because somebody has to have access to the people, ask the right questions, understand the answer, verify what matters and publish it while it is still useful.

The opportunity becomes stronger again when that knowledge is current. If an organisation can speak to the people actually dealing with a change now, rather than publishing another synthesis of what was already online, it has the chance to add information to the wider knowledge environment rather than merely rearrange it.

This is where journalism becomes a GEO capability

That is why I keep coming back to journalism. A journalist does not normally start by asking which existing page should be made more extractable. The starting questions are more basic and more demanding: who knows something, what has changed, what is still unclear, what evidence exists, who disagrees, and what do I need to ask before I can explain this properly?

I have explored the practical version of this in Becoming the GEO journalist. A weak answer can sometimes be fixed by renewing the existing page, and sometimes stronger evidence already exists elsewhere and simply needs to be routed into the answer. But when the organisation genuinely does not know enough, the next intervention is not another writing brief. It is reporting.

That does not mean first-party material becomes a magic GEO tactic. A customer quote does not mechanically increase citation probability, and an interview is not valuable simply because a human took part. The stronger claim is editorial: specialised, attributable evidence can make an answer more specific, harder to reproduce and more useful than another synthesis of public material. Whether that subsequently improves AI visibility is something to measure rather than assume.

I am already testing the more basic premise in customer storytelling

One reason to be cautious with this argument is that I do not yet have a clean case study showing that this approach improves GEO performance. What I do have is a live editorial test of the underlying knowledge problem.

In customer-story work, the conventional unit of value is the finished case study. An interview is commissioned, the customer explains what happened, the writer selects the strongest material, the story is approved and published, and the project is complete. That workflow produces a useful asset, but it also throws away a surprising amount of intelligence.

The interview may have revealed the story spine, proof points, useful customer language, an implementation difficulty, an unexpected outcome, an evidence gap, a theme that recurs elsewhere or a question that the interviewer wishes they had asked. Most of those things are not needed in the final article, so they normally remain in the transcript or disappear from the working process altogether.

I am now testing a different approach in live B2B customer-story work: after the story is completed, capture some of that surrounding intelligence in a structured evidence register. The fields are deliberately practical — strongest proof points, useful approved quotes, evidence gaps, questions we wish we had asked, themes supported and editorial lessons — because the aim is not to archive everything. It is to preserve the parts most likely to improve future work.

The immediate test is not whether an AI system cites the resulting customer story more often. It is whether the next interview starts smarter because the previous one left behind usable knowledge. Can recurring gaps change the questions we ask? Can evidence from several stories reveal a pattern worth investigating? Can a proof point captured for one deliverable strengthen a different answer later without somebody having to rediscover it from scratch?

That is a much more modest claim than saying the system has proven a GEO advantage. It is also more useful evidence for this argument, because it tests whether editorial work can become cumulative rather than disposable.

The format is not the strategy

This changes how I think about the familiar question of which content formats organisations should produce for GEO. I am less interested in whether AI systems prefer articles, FAQs, video, research or customer stories than in what each format can contribute to a question the organisation needs to answer better.

A customer story can provide proof that something happened in practice. Research can establish scale or identify a pattern. An expert interview can explain mechanism and nuance. A podcast can expose tacit knowledge that would never appear in a formal brief, while video can preserve the voice and context of someone who has actually done the work. A product page can explain capability, and independent PR coverage can show that an argument or piece of evidence has travelled beyond the organisation’s own channels.

The point is not to create another GEO workflow with its own named sequence. I already have an Editorial Intelligence cycle — Generate → Capture → Connect → Synthesise → Shape → Activate → Learn — and the GEO-informed thought leadership guide uses that same cycle rather than inventing another one.

GEO changes some of the inputs. AI answer environments can reveal questions, weak visibility, missing evidence and competing sources. Specialist interviews, customer conversations and research can generate stronger evidence. Technical work can make the resulting knowledge easier to retrieve. Performance and citation signals can then contribute to the Learn stage. The underlying editorial system does not need a fourth loop to describe that.

The most important step may happen after publication

The more interesting change is what happens after a deliverable is finished. A customer interview produces a case study, a podcast produces an episode, a research project produces a report, and most content systems then move on to the next item on the calendar.

Yet each activity may have generated intelligence beyond the published asset. A customer may have described an unexpected implementation problem, an expert may have introduced a useful distinction, a guest may have answered a question more clearly than the organisation ever has, or three different interviews may reveal a recurring pattern that nobody was looking for when the individual pieces were commissioned.

If that intelligence is captured and connected, the organisation starts building memory. The next customer interview can ask a better question because of what previous interviews revealed; a future podcast can pursue a contradiction surfaced by an earlier guest; a research brief can test an emerging pattern rather than generate another disconnected statistic.

This is one reason I think GEO is potentially more important as a reason to improve the editorial system than as a new content format. AI visibility gives teams another way to observe how their knowledge is being encountered, but the larger gain comes when that observation changes what they investigate, capture and publish next.

Brand, PR, performance and GEO should compound rather than compete

I developed the broader connection in A guide to GEO-informed thought leadership. The useful insight for me was that GEO becomes less interesting when it is treated as another silo and more interesting when it helps brand, editorial, PR, performance and thought leadership work on the same body of knowledge.

Brand helps decide what the organisation wants to become credibly associated with. Thought leadership develops that intellectual territory over time rather than treating every article as a fresh start. Editorial finds the questions, people and evidence worth pursuing, while PR helps the resulting expertise travel through external conversations and trusted third-party environments.

Performance then has a more useful job than reporting whether the last asset hit a traffic target. Search behaviour, AI citations, referral patterns, reader response, journalist questions, sales conversations and other signals can show which parts of the territory are resonating, where the organisation is absent, which claims travel without their intended interpretation and what people still want to know.

GEO can contribute to that learning system without becoming a separate methodology. It can expose answer gaps and visibility gaps; technical work can make strong material easier to retrieve; the response can then feed back into the same Editorial Intelligence cycle. The important question is not whether every content type has been made “GEO-friendly”. It is whether the organisation is becoming better at knowing, proving and explaining something useful.

The podcast idea is still a hypothesis

This is where I need to distinguish more clearly between what I have already tested and what I think should work. I have argued elsewhere that organisations should get more value from video by treating interviews and recordings as sources of reusable expertise rather than isolated assets. That remains a hypothesis I want to test more deliberately in live work, not something I can yet present as established practice.

The test would be straightforward. Before an interview or podcast, identify a small number of important questions where the organisation’s current evidence is weak. Use the conversation to investigate only the gaps the guest is genuinely qualified to address. Afterwards, capture the strongest attributable explanation, note what remains unresolved and see whether the material improves an existing answer or changes the next brief.

If that works, the value of the podcast is not that video has some inherent GEO advantage. It is that access to a specialist produced knowledge the organisation did not previously possess, and the editorial system remembered enough of it to improve future work. If it does not change any subsequent decision, then the knowledge-system claim is weaker than I currently think it is.

That is a better test than treating the podcast itself as proof.

The counterargument matters: human does not automatically mean useful

There is an easy way to overstate all of this. Real people can produce generic content too. An executive repeating approved messaging is not more valuable simply because there is a face attached to it, and a customer anecdote is not automatically representative or trustworthy because it is first-hand.

Specialisation matters because the person needs to know something relevant, and editorial judgement matters because somebody has to test what the evidence supports. Recency matters when the subject is changing quickly, but current does not automatically mean accurate. Original research can be weak, customer quotes can be cherry-picked and thought leadership can become elaborate brand theatre.

The technical side also remains necessary. Strong knowledge that cannot be crawled, understood, attributed or connected is needlessly hard to discover, while useful expertise buried inside an inaccessible video may effectively be invisible to many retrieval systems and human readers alike.

So I would not replace technical GEO with editorial romanticism. I would put them in the right order: build or uncover something worth retrieving, then make it retrievable.

My bias is part of the test

I am still aware that this is a convenient conclusion for me. It moves GEO towards journalism, customer insight, editorial strategy, thought leadership, evidence systems and narrative work — exactly the territory in which I already feel comfortable.

If the evidence ultimately shows that AI visibility is mostly won through technical execution and large-scale optimisation, then my preferred version of the discipline will not become correct because I like it more. Equally, I do not think editorial teams should ignore a potentially important strategic role simply because the argument happens to suit their existing skills.

The useful next step is not another GEO framework or another theoretical loop. It is more applied evidence. Does structured memory from customer interviews improve future questions and evidence choices? Can an AI visibility gap change a real brief before publication? Can an expert conversation produce a stronger answer than the organisation previously had? Do PR, audience and retrieval signals become more useful when they feed the same editorial system?

Those are questions worth testing in live work rather than settling through positioning language.

For now, I keep returning to one distinction. Technical GEO can help get an organisation’s knowledge into the retrieval system, but editorial intelligence determines whether there is anything distinctive there worth retrieving.

Most GEO advice starts with the engine. I think editorial teams should start with what the organisation knows — and with the specialised people who can help it learn what it does not know yet.