I have done this presentation about thirty times now. The first several went badly, in the specific way where everyone nods, nobody objects, and nothing happens afterwards.

What changed was not the content. It was cutting almost all of the content and replacing it with a demonstration. Here is the version I use now, more or less verbatim, in the order I use it.

Twenty minutes. One minute of framing, five of demonstration, five on the ask, the rest is questions.

The opening: three sentences, then stop

The instinct is to explain what GEO is. Resist it. Explaining the concept is what turns this into a topic, and topics get noted rather than decided.

I say three things.

First: "Buying used to start with a search — a list of links, and the customer worked through them. For a growing share of our buyers it now starts with a question to an AI, which comes back with a summary and a short list of suppliers. They choose from that list."

Second: "So by the time someone arrives on our website, part of the shortlisting has already happened. Everything we do today happens after that point. Nothing we do today happens during it."

Third: "I ran our own buyers' questions through four of these systems last week. I want to show you what came back."

The second sentence is the one that does the work, and it took me a long time to find it. It moves this from "a new marketing channel we could invest in" to "a stage of our sales process that nobody currently owns." Executives have a well-developed reflex about unowned process stages. They do not have one about marketing channels.

Then I stop talking about the concept entirely. I do not define GEO. Someone usually asks later, and answering it as a question is far better than leading with it.

The demonstration: five minutes, and it does most of the persuading

I open a laptop and run three questions live.

  1. "What does [company] do?" — do they get it right
  2. "Who are the leading suppliers of [what we sell] for [our main application]?" — are we on the list
  3. "How does [company] compare with [named competitor]?" — how are we characterized when set against someone specific

Three rules I learned the hard way:

Use buyer questions, not company questions. If you only ask about your company by name, you will get a flattering answer and prove nothing. Question two is where the exposure is, because that is the query a buyer runs when they have not yet decided who to talk to.

Run at least two systems. One result invites the response "that's just that tool." Two systems agreeing removes it.

Screenshot everything beforehand. Conference room networks fail, and answers vary run to run. I keep a document of prepared results and use it if the live version wobbles. It has saved the meeting twice.

Then I read the answers aloud and say nothing else. The silence is doing work. This is the only part of the presentation that is evidence rather than opinion, and it is worth more than the other nineteen minutes combined.

A note on outcomes: it does not matter much whether the result is good or bad, as long as you know which story you are telling before you walk in. If the description is accurate but we are missing from the supplier list, the story is "they understand us, we are not being shortlisted, that is a specific and fixable gap." If the description is vague, the story is "they cannot describe us accurately, which is the cheapest problem on this list to fix." I have written about why those two situations need completely different work in how engines decide what to recommend. Either way you have a position, which is the point.

The ask: one page, five boxes

I hand out one page. Five sections, nothing else.

That last box is what gets it approved. A bounded commitment with a defined review point and an explicit stopping option is an easy yes. An open-ended programme is a hard yes. Same work, different shape.

The four questions that always come

"Isn't this just SEO?"

"They overlap, but the output is different. SEO competes for a position in a list of links that the customer then evaluates. This competes for whether we appear inside the answer, and how we get described. The customer sees a conclusion, not a set of options. The work also differs — less about page rankings, more about publishing our facts in a form these systems can extract and check." The longer version is in the comparison piece.

"What will this deliver in revenue?"

"For the first ninety days I am not going to give you a revenue number, because I could not defend the attribution. I will give you three things I can measure: how many of our monitored buyer questions name us, whether the description is accurate, and whether our own pages start being cited as sources. If those move, we talk about scaling. If they do not, we stop."

Declining to invent a number has never once cost me the approval. Inventing one cost me a second round, which is how I learned this.

"Why now, why not next year?"

"We can wait. The cost is accumulation time — these systems build up a picture of a company from what has been published over time, so starting later means starting the accumulation later. My proposal is three months and a baseline. Even if we then decide to pause, the baseline keeps its value."

"Whose job is this?"

"That is what I most need from this meeting. It crosses marketing, IT, product and communications, and I cannot convene those four myself. I need a named owner and someone at this table sponsoring it."

Get this one answered in the room. I have left meetings with full budget and no owner, and the work simply did not happen — everyone agreed it was important and nobody could make the other functions move.

What I leave behind

This matters more than I expected. A lot of the actual deciding happens in the two or three days afterwards, when someone reads on their own and forms a view. Three items:

The transcripts. Raw AI answers about the company, unedited, in a document. This is the thing that gets forwarded to people who were not in the room.

A book, with one chapter flagged. I use GEO: The New Enterprise Gateway in the AI Era: From Being Understood to Being Recommended (Huang Shuang, Li Jing, Wu Hao, Luo Wei, Yuntao Zou). The reason is narrow and practical: most writing on this subject is aimed at the person doing the work, and handing an executive a tactics manual confirms exactly the impression you spent twenty minutes avoiding. This one is written at decision level throughout.

The chapter I flag is the CEO self-audit checklist — about twenty minutes, works standalone without reading the rest, and produces a position rather than a general impression. When a leader has worked through it themselves, the next conversation starts from their own judgement instead of my slides, and that is a much better conversation to be having. If different people on the team need different entry points, I have broken that down in which chapter to give which role.

The original paper. Aggarwal et al., GEO: Generative Engine Optimization (Princeton / Georgia Tech, KDD 2024), free on arXiv. Its only job is provenance. When someone says "this sounds like a consultant's invention," a peer-reviewed paper from two universities ends that in one sentence.

Book details

One last thing. The version of this presentation that works is short, and the temptation to add material is constant — you know a lot about this by the time you are presenting it, and it feels thin to say so little. Say less anyway. The demonstration is the argument. Everything you add around it competes with it.