Ninety seconds that ended the argument

Disclosure: this post contains affiliate links. If you sign up through them I earn a commission at no extra cost to you. I only write about tools I run on my own properties.

Ninety seconds that ended the argument

A client asked me last month whether they should pause their SEO retainer and start "doing the AI thing" instead. Fair question — they'd read three newsletters telling them search was dead and that Generative Engine Optimisation was a new discipline they were already behind on.

So I did something simple, live, on the call.

I opened ChatGPT and typed the exact question their customers ask. Not their brand name — the unbranded, commercial question. Something close to "best airport transfer company in Manchester."

They weren't mentioned.

The three companies that were? All ranking in the top five in Google for the same query.

That wasn't a coincidence, and it reframed the entire conversation. They didn't need a new discipline. They needed to rank, and they needed to be able to see whether it was landing.

There is no separate AI channel to win

Here's the part the "SEO is dead" crowd consistently skips.

When a language model answers a question with buying intent — about a specific service, in a specific place, right now — it is not reciting something it memorised during training. Training data is frozen and generic. Commercial answers require current, local, specific facts.

So the model retrieves. It runs searches, pulls live web results, and reasons over what comes back. And AI Overviews sit directly on top of Google's existing index, drawing from the same corpus of ranked pages you've been optimising for all along.

Which means the crawlable, rankable web is still the substrate. AI search is a new interface on top of the old inventory, not a new inventory.

Once you internalise that, most of the panic evaporates and the practical implications get clear.

What "GEO work" actually looks like

When people ask me what optimising for AI search involves, the honest answer disappoints them, because it looks almost exactly like good SEO. Here's the list, with the reason each item matters to a retrieval system specifically.

1. Clean, complete structured data

Schema markup is how a machine knows, without inference, that you are a business, of a certain type, in a certain city, offering named services, at price ranges, with a rating and opening hours.

A model assembling an answer about local providers has to resolve ambiguity fast. Explicit structured data removes that work. Sites without it require the model to guess from prose, and guessing is where you get skipped in favour of a competitor who made it easy.

LocalBusiness, Service, FAQPage, Article and Organization markup. Correctly nested, validated, present on the pages that matter.

2. Clear entity definitions

Go and read your own homepage as if you were a machine trying to extract facts.

Most service-business homepages are adjective soup. "Premium." "Trusted." "Seamless." "Your journey, elevated." There is almost nothing extractable — no statement of what you do, where, for whom, at what price, since when.

Compare: "We operate chauffeur-driven airport transfers between Greater Manchester and Manchester, Liverpool and Leeds Bradford airports, for corporate accounts and private travellers, since 2011."

That second sentence can be lifted verbatim into an answer. The first cannot be lifted at all. This is the cheapest, highest-leverage change most sites can make, and it costs an afternoon.

3. Direct answers to direct questions

A model that wants to cite you needs a sentence it can quote. If your answer to "how much does a Manchester airport transfer cost?" is implied across paragraphs four through seven of a page about your company values, it functionally doesn't exist.

State the question as a heading. Answer it in the first sentence beneath. Elaborate after. This is good writing anyway — it just happens to also be machine-readable.

4. Genuine topical depth

One post on a subject signals nothing. Twenty interlinked pieces across a cluster, pointing at a pillar page, signals that you are the reference on that subject.

That has always been how topical authority worked for ranking. It turns out to be even more decisive for retrieval, because a system choosing between sources will favour the one with demonstrable depth over the one with a single thin page.

This is exactly why the content volume question and the AI visibility question turn out to be the same question. Depth is the mechanism for both.

5. Being cited elsewhere

Corroboration matters. A claim that appears only on your own site is weaker than one that appears on your site and is referenced by others. That's true for ranking and it's true for a model weighing which sources to trust.

Which folds link acquisition into the same picture rather than making it a separate project.

Look at that list again. Schema, entity clarity, direct answers, depth, citations. Every item is something a competent SEO has been doing for a decade. Nothing on it is new.

So what actually changed? Measurement.

This is the real break, and almost nobody talks about it because it's less exciting than declaring a discipline dead.

You have Search Console. You know your impressions, your clicks, your average position, your query mix, your page-level performance. Fifteen years of tooling has been built to tell you how you're doing in the ten blue links.

Now answer this: do you know whether your brand appears in AI answers at all?

Almost nobody does. And in my experience the people who assume they do assume optimistically, because they've made the same mistake:

They typed their own brand name into ChatGPT, watched it produce a flattering paragraph, and concluded they were visible.

That's not visibility. That's a model retrieving information about a company you explicitly named. Nobody searching for a supplier types your brand name — that is the entire definition of acquisition. The question that matters is whether you're cited when someone asks the unbranded, commercial question. And until recently there was no clean way to answer that beyond typing prompts into four chatbots and screenshotting results.

Find out where you actually appear

Track brand citations across ChatGPT, Perplexity, Claude and Gemini — which prompts trigger a mention, and how it trends.

See AI visibility tracking →

Closing the measurement gap

Across the 185-odd properties I look after, AI visibility went from a novelty question to a standing agenda item on client calls sometime last year. I needed a number, not an anecdote.

That's the piece I've found genuinely useful in BabyLoveGrowth.ai. Most of the platform is content automation — keyword research, a 30-day plan, articles written and published to your CMS with schema and internal links, plus contextual backlinks. That's the headline, and it's why small retainers become viable.

But the feature I keep opening it for is the AI visibility tracking. It monitors brand citations across ChatGPT, Perplexity, Claude and Gemini, shows which prompts trigger a mention, and trends that over time.

Being able to tell a client "here is exactly where you appear in AI answers, here are the specific prompts, here's the direction it's moving" transforms the conversation. It stops being a philosophical debate about whether AI search matters and becomes a metric that either moves or doesn't.

That's the whole job of a good report: convert an argument into a number.

Run this audit yourself this week

You don't need to buy anything to start. Do this manually first — it takes an hour and it will tell you more than any article.

Step 1: Write five unbranded prompts

Not "tell me about [your company]." The questions a buyer with no knowledge of you would type. For a transport operator: "best executive car service near Heathrow," "how much does an airport transfer from Manchester cost," "reliable chauffeur company for corporate accounts in the North West."

Step 2: Run all five across four systems

ChatGPT, Perplexity, Gemini and Google's AI Overviews. Twenty prompts total. Log the result in a spreadsheet: which brands were named, in what order, and which sources were linked.

Step 3: Cross-reference with Google

Search the same five queries normally. Note who ranks in the top five.

Then compare the two lists. In nearly every audit I've run, the overlap is substantial — usually seventy percent or more of AI-cited brands are also top-five organic results. That overlap is your entire strategy, made visible.

Step 4: Identify your gap

You'll be in one of three positions:

• Ranking and cited. Good. Protect it and expand the cluster.
• Ranking but not cited. Usually a schema, entity clarity or answer-formatting problem. Your page is there but not extractable. This is the most fixable position and the fastest win available.
• Neither. You don't have a GEO problem. You have an SEO problem. Go and fix that first — depth, links, technical foundations. AI citation follows.

Step 5: Repeat quarterly

One reading is a data point. Four readings is a trend. If you're doing this manually, calendar it — and if you'd rather not run twenty prompts by hand every quarter, that's precisely the job automated tracking does.

The five prompt types your buyers actually use

When people build their prompt list for the audit above, they usually write five variations of the same query. That wastes the exercise. Buyers ask structurally different kinds of question at different stages, and you can be visible for one type and invisible for another.

1. The recommendation prompt

"Best [category] company in [place]" — the highest-value and most competitive. The model returns a shortlist, and being on it is worth more than any single ranking position, because a shortlist of three is a far narrower funnel than a page of ten results.

What earns inclusion: topical depth, consistent mentions across multiple sources, clear structured data, and being findable in organic results for the same query.

2. The comparison prompt

"[Competitor] vs [Competitor]" or "alternatives to [Competitor]" — enormously valuable and frequently overlooked. If your competitors have comparison content and you don't, models will describe your category using their framing, with their criteria, and you'll appear as a footnote if at all.

What earns inclusion: honest comparison content on your own site, and being referenced in third-party roundups.

3. The specification prompt

"How much does [service] cost in [place]" or "how long does [process] take" — factual, extractable, and the easiest to win. Models want a concrete number with context. If you publish a clear pricing explanation and your competitors publish "contact us for a quote," you get cited and they don't.

This is the single fastest win available to most service businesses, and it's blocked almost entirely by a reluctance to publish prices.

4. The problem prompt

"What do I do if [problem]" — top of funnel, high volume, lower intent. Worth covering because it establishes topical depth, and depth is what qualifies you for the recommendation prompts.

5. The trust prompt

"Is [your brand] reliable" or "[your brand] reviews" — this one runs after someone has already heard of you, and it's the one where third-party signals dominate entirely. Your own site has limited influence. Review platforms, directories and independent mentions do the work.

Run at least one of each type in your audit. The pattern of where you appear and where you don't will tell you what to fix far more precisely than an overall visibility score.

You rank but you're not cited. Now what?

This is the most common and most fixable diagnosis, so it deserves its own treatment. Your page is in the top five, the model retrieved it, and still didn't use it. Work through these in order.

Your answer isn't extractable

The single most common cause. The information is on the page but distributed across several paragraphs, hedged, or buried under preamble. Rewrite so the direct answer is the first sentence under a heading that matches the question.

Your schema is missing or malformed

Run the page through a structured data validator. Missing markup is common. Malformed markup that silently fails validation is more common still, and worse, because you think you've done it.

Your content is thin relative to competitors

If the pages being cited are 2,000 words of genuine depth with original detail and yours is 600 words of summary, the model is making a reasonable choice. Depth is the fix, and it's slower than the others.

You have no corroboration

If a claim appears only on your own site and nowhere else, it carries less weight than one echoed by independent sources. This is another reason link and mention acquisition matters here rather than being a separate workstream.

Your entity is ambiguous

If your business name is generic, or your service area unclear, or your category unstated, the model may not be confident enough to name you. Fix the About page, the schema, and the homepage first paragraph.

You're blocking the crawlers

Check robots.txt. Some sites block AI user agents deliberately, some inherited a block from a template or a plugin default and have no idea. Both are fine as decisions; only one is a decision.

Three things people get wrong about GEO

"It's a completely new skill set"

It isn't. It's SEO with the addition of structured data discipline and answer-first formatting. Anyone selling you a GEO course that doesn't spend most of its time on things you'd recognise from 2018 is selling novelty.

"AI search will kill organic traffic"

Zero-click behaviour is real and worth planning for. But the response isn't to abandon organic — it's to recognise that being the cited source in a zero-click answer is now itself a brand impression with commercial value. And that citation is earned by the same work as ranking.

Plan for a world where some queries never produce a click, and where being named in the answer is the win. Then measure the thing that actually correlates with revenue for you — enquiries, calls, bookings — rather than sessions alone.

"We should write content specifically for AI"

Please don't. Content written to game retrieval reads exactly like content written to game rankings did in 2012, and ages just as badly. Write for the person. Format so a machine can extract it. Those two goals are almost entirely aligned — clear answers, clean structure and real substance serve both.

Frequently asked questions

Does ranking in Google guarantee I'll be cited by ChatGPT?
No — it substantially raises the probability. Retrieval draws heavily on ranked results, but the model still selects among them based on how clearly your page answers the specific question and how extractable that answer is.

Is GEO different from SEO?
The work overlaps almost completely. The differences are emphasis: more weight on structured data and answer-first formatting, more attention to entity clarity, and a different measurement layer.

How do I know if AI systems can even read my site?
Check your robots.txt isn't blocking AI crawlers if you want to be included, validate your schema, and confirm your key content isn't rendered exclusively client-side in a way that makes extraction unreliable.

Should I block AI crawlers?
That's a genuine business decision with two defensible sides. Publishers protecting subscription revenue have real reasons to block. Service businesses that want to be recommended almost always don't. Decide deliberately rather than by default.

Which AI platforms should I track?
ChatGPT, Perplexity, Gemini, Claude and Google's AI Overviews cover the overwhelming majority of current usage.

How long until this shows results?
The same timescale as SEO, because it is SEO. Ninety days for early signal, six months for meaningful movement on anything competitive.

The bottom line

SEO didn't get replaced in 2026. It got a second surface.

The same content, entities and links that earn you position three now also earn you a citation in a generated answer — for free, as a side effect, whether or not you ever called it GEO.

What you're missing isn't a new strategy. It's the instrumentation to find out whether the strategy you already have is landing on that second surface.

Go and measure it. You may be doing considerably better than you think, or considerably worse. Either way, that's worth knowing before you tear up a retainer on the strength of a newsletter headline.

Rank in Google. Get cited by AI.

Automated content and links, plus citation tracking across ChatGPT, Perplexity, Claude and Gemini.

Try BabyLoveGrowth.ai →

Written by Sunil, founder of ElevateCode Digital. Building in the ground-transport software and web design space since 2008, with 250+ operator deployments across 50+ countries, and currently managing SEO across 185+ properties in the UK, US, Australia and UAE. Affiliate links are marked; I only recommend tools I run on my own properties.