Being cited by ChatGPT is the new SEO

For twenty years, being visible meant one thing: appearing on the first page of Google. Then your clients changed their habit. They no longer type three words into a search bar. They ask an AI a full question, and they read a single answer.

This change is brutal for an SME. On Google, page two still existed. In an AI answer there is no page two. Either you are in the answer, or you do not exist.

What AIs actually read

Models do not rank keywords, they repeat what is clear, structured and verifiable. A company whose site says precisely what it does, for whom, where and at what price is ahead of the one stacking slogans. AIs cite sources that answer the question, not those that shout loudest.

Three things matter most: pages that answer real client questions, consistent data everywhere your company appears, and concrete proof, numbers, cases, addresses, rather than adjectives.

A Geneva restaurant, as the AIs see it

Take a case we meet often, anonymised but faithful in its mechanics. A neighbourhood restaurant in Geneva, seasonal cooking, full room at weekends, quiet weekdays. The owner runs the test: he asks ChatGPT where to eat market cuisine near the station. The answer names three places. Not his. Not because his food is worse, but because his site says welcome and serves the menu as a scanned PDF.

What the AIs found at the other three: a page describing the cooking in plain words, a price range, address and hours in text, a mention in a local article, a Google listing consistent with the site, reviews telling concrete details. The machine invented nothing. It assembled what was readable, and he was readable nowhere.

The fix is not heroic. The menu moves from PDF to text, dishes and prices included. One page answers the real questions: lunch menu, terrace, groups, vegetarian options. Opening hours become identical everywhere, site, Google listing, directories. One page says who cooks, with a name and regional suppliers.

A few weeks later the restaurant shows up in some of the answers. Not all of them, and nobody can promise otherwise. The lesson fits in one sentence: the machine can only cite what it can read. As long as your reality is not written down somewhere, it does not exist for it.

Where to start

We call it GEO, generative engine optimization. The work starts with an inventory: what do ChatGPT, Perplexity and Gemini answer today when asked for an agency, a craftsman, a service in your region? If the answer does not mention you, it mentions a competitor.

The good news: almost nobody in French-speaking Switzerland does this work yet. Tomorrow's visibility is built now, while the field is open.

And what about Google

The question always follows: is classic SEO still worth doing? Yes, for a mechanical reason. Part of the AIs rely on search indexes to find their sources: a well-ranked page has more chances of being read, therefore cited. The groundwork you may already have paid for, clean structure, fast pages, useful content, is not lost. It has become the foundation of a second floor.

Local search has not moved a centimetre. People still look for a plumber open on Saturday or a table for tonight, and the map remains the first reflex for anything with an address. Your Google listing feeds both worlds at once: classic results and generated answers. Taking care of it is probably the best-invested hour of this whole project.

Structured data counts too. Hours, prices, address, reviews: when this information is properly marked up in your site's code, it becomes unambiguously readable for every machine, engines and models alike. It is technical, invisible work, half a day for an SME site, and it serves both channels for years.

Third-party mentions weigh more than before. An article in the local press, an entry in a professional directory, a talk at an industry event: every written trace somewhere other than your own site is an independent source confirming your existence. Models cross-check sources; the more they agree, the more citable you become.

What really changes is the measurement. Yesterday you tracked positions on keywords. Today you also need to track a presence inside answers, which is done by hand, with the quarterly routine described below. The two measurements tell different stories, and a serious SME watches both without worshipping either.

A word on budgets: none of this requires doubling the spend. The technical base serves both channels, useful content serves both channels, the Google listing serves both channels. What gets added is mostly discipline, not invoice lines.

In plain terms: GEO does not replace SEO, it stacks on top of it. Abandoning Google because the AIs are coming would be as absurd as ignoring Google because the paper directory still existed. Your clients' habits add up; your visibility has to add up the same way.

The method, step by step

Step one, the inventory. Ask the AIs some twenty questions a real client would ask, in everyday words, with your region named. Note who gets cited, how, and which sources the answer mentions. Do it on ChatGPT, Perplexity and Gemini, because they do not read the same sources. That record is your measurable starting point.

Step two, the diagnosis. Compare what the answers say with what you actually sell. The gaps are your work list: a service missing from the site, a price nowhere to be found, an address that differs between two directories, an offer described with slogans instead of facts.

Step three, the factual base. Name, address, hours, phone, services: identical everywhere your company appears. It is thankless, it is tidying up, and it is what makes the machine trust you. One inconsistency between two sources is enough to make it move on to the next competitor.

Step four, the pages that answer. Take the questions from step one and write one page per frequent question. The question as the title, the answer in the first lines, the proof after. No detours, no preamble about your passion.

Step five, the proof. Cases described with numbers, photos of real work or real dishes, team names, mentions in the local press. Models favour what can be verified somewhere other than your own site.

Step six, the routine. Replay your twenty questions every quarter and compare with the previous record. What moved tells you what works. Visibility in AI answers is maintenance, not a one-off project.

The mistakes that cost dearly

First mistake: writing for robots. Keyword stuffing died with the old SEO. Models summarise meaning, not density. A text unreadable for a human is unusable for the machine, and it drives away the only reader who pays.

Second mistake: publishing fifty generic articles produced in one afternoon. Models pick up what sets a source apart. Interchangeable text adds nothing, it even dilutes what was specific about you. Ten right pages beat a hundred lukewarm ones.

Third mistake: buying mentions or reviews. The inconsistencies show sooner or later, and what is online stays online. Your reputation with machines is built like your reputation with people: slowly, and it is lost fast.

Fourth mistake: aiming to be first everywhere. An SME will not win the generic question asked from Zurich. It can win the precise, local questions, the ones where the answer only cites two or three names. That is exactly where clients are won.

What you can start this week

Open ChatGPT and ask five questions your clients ask you on the phone. Copy the answers into a dated document. That is your point zero, the one that lets you measure progress instead of assuming it.

Then fix the most visible data: hours, address, phone, services, identical on the site, the Google listing and the directories. One hour of work, an effect that lasts for years.

Finally pick one frequent question and turn it into one clear page. Then another next month. Consistency beats the sprint. In a year, when your competitors get started, you will be twelve pages ahead with a history the machine already knows.