What is Generative Engine Optimization (GEO)?

Generative Engine Optimization (GEO) is the practice of making sure your business appears, accurately, when someone asks an AI assistant such as ChatGPT, Gemini, Claude or Perplexity for a recommendation in your market. It is the same problem search visibility solved fifteen years ago, in a new place — and most businesses have not looked yet.

Why does GEO matter now?

Because the question your customer asks has changed shape. For twenty years, someone looking for a supplier typed a few words into Google and received a list of ten links. They did the choosing. Today a growing share of those people type a full sentence into an AI assistant — "who's a good bookkeeper in Johannesburg for a small construction business?" — and receive a short, confident answer naming two or three options.

That is a completely different competitive situation. A list of ten links has room for you. A generated answer naming three businesses usually does not. The shortlist is built before the customer has seen a single website, and if you are not in it, you are not in the running — you never even register as an option that was rejected.

Most owners have never checked whether they appear. That is the honest starting point for almost every business I speak to.

How is GEO different from SEO?

SEO optimises for a ranked list of links. GEO optimises for being included in a generated answer. The mechanics differ enough that doing one well does not automatically get you the other.

The old model and the new model, side by side:

• Old model: business → website and content → Google → search result
• New model: business → entity signals, content and reputation → AI systems → generated recommendation

Notice what changed in the middle. In the old model the website was the thing being ranked. In the new model the website is one input among several — alongside directories, review sites, mentions elsewhere on the web, structured data, and whatever the model absorbed during training. You are no longer optimising a page. You are optimising what the internet collectively appears to know about your business.

Why do AI answers keep changing?

This is the part that surprises people most, and it has real consequences for how you measure.

Generative systems do not look up a fixed ranking. They produce an answer, and there is inherent variability in that process. Ask the same question five times and you can get five subtly different answers — different businesses named, in a different order, sometimes with one dropping out entirely.

Personalisation adds another layer. If you are logged in, the assistant may draw on your previous conversations, your stated location, or your account history. Your own business will often appear when you ask about your own market — not because you are genuinely visible, but because the system knows it is you asking.

This is exactly why a single check is close to worthless. I have run repeated, identical queries about a service business and watched it come back ranked first in some answers and be omitted entirely from others in the same session. One run told me almost nothing. Ten runs told me the actual picture: it appeared roughly half the time, which is a very different problem from "we're invisible" and needs a very different response.

So the correct unit of measurement is not a position. It is a rate — how often you appear across repeated runs of the questions that matter.

What signals do AI systems draw on?

Nobody outside those companies can give you a definitive list, and anyone claiming otherwise is guessing with confidence. But the pattern across observed behaviour is consistent enough to work with:

• Clear entity facts. What the business is, where it operates, who it serves, stated plainly and consistently in text — not implied, not only in images, not buried in a PDF.
• Consistency across sources. The same business name, contact details and description on your site, your listings and your profiles. Contradictions weaken confidence.
• Structured data. Schema markup that states, in machine-readable form, that this organisation exists, what it does and where.
• Third-party corroboration. Being mentioned, reviewed or listed somewhere other than your own website. Self-description alone is weak evidence.
• Content that answers the actual question. Pages written as direct answers to real questions, rather than marketing copy that circles the subject.
• Crawlability. If your key claims only appear after JavaScript runs, or your robots file blocks AI crawlers, you have made yourself hard to include.

That last one catches people out. I have seen businesses spend money on visibility while quietly blocking the crawlers that would have to read them.

What does GEO work actually involve?

In practice it is a loop, not a one-off fix:

1. Establish which queries matter. Not the words you would use — the words a customer would use. "Who can help me sort out my invoicing?" rather than "business process automation services".
2. Measure how often you appear. Run each query repeatedly, in clean sessions, and record the rate.
3. Identify who is recommended instead. This is usually the most uncomfortable and most useful part. The competitors that show up are telling you what the systems currently consider credible.
4. Work out which signals are missing. Compare what exists about you against what exists about them.
5. Improve the signals. Entity facts, structured data, content architecture, third-party presence.
6. Monitor over time. Because the systems change, and because the only proof that it worked is the rate moving.

How long does GEO take to show results?

Slower than automation work, and anyone promising otherwise is selling. The measurement itself takes days. Fixing entity and structured-data problems is quick. But the systems need to re-encounter your business, and third-party signals accumulate at their own pace. Expect to measure meaningful change over roughly 30 to 90 days, and expect it to be a rate shifting rather than a switch flipping.

Is GEO worth doing for a small business?

It depends on whether your customers ask for recommendations before they buy. For a professional service, a supplier, a specialist trade — the answer is usually yes, because that is exactly the kind of question people bring to an assistant. For a business that trades on foot traffic or on relationships that never begin with a search, it matters much less.

That is a judgement call about your specific market, and it is worth making honestly before spending anything. The failure mode I see most often is not businesses ignoring GEO. It is businesses buying "AI visibility" as a product without first checking whether their customers are asking AI anything at all.

Teeanu "Teejay" Thomas

Founder of Phantom Ops (Pty) Ltd, an AI transformation partner in Johannesburg, South Africa. Registered POPIA Information Officer with South Africa's Information Regulator. Works with owner-run South African businesses on where AI creates measurable leverage.