How to get your business named by ChatGPT and Gemini
What actually influences whether an AI assistant names your business, what no supplier can control and how to test it without fooling yourself.
Why your name does not come up
A buyer types a question into an assistant and asks for the best supplier of something you sell. Three names appear and none of them is yours. That is the whole problem and it is becoming a normal way for people to shortlist.
Understanding the mechanism matters before spending anything on it. An assistant answers either from what the model absorbed during training or from pages it retrieves at the moment of the question or from a mixture of both. Nobody outside the companies that build these systems can edit the first. The second is a search and retrieval step and it draws on sources that can be influenced. That narrow gap is the entire opportunity.
What this excludes and what it costs
Nothing anyone sells can insert your business into a model's training data. Nothing can place you at a position in an answer because there is no ranked list to hold a position in. Two people asking the same question in the same hour can receive different names depending on wording region account history and which version of the model is serving them. Any supplier describing a rank or a guaranteed mention is describing something that does not exist.
What it costs is content work and reference hygiene. Somebody has to write genuinely useful pages. Somebody has to correct your details across directories review sites and industry listings. Technical work is needed so retrieval tools can read your pages at all. Fees for an AEO and GEO engagement sit on the service page and carry GST at 18%. None of it buys an outcome and the honest version of this work is sold as effort rather than placement.
Who should do this and who should not
It suits businesses whose buyers research in sentences before deciding. Software services education healthcare services specialist manufacturing travel and anything where people ask for a comparison or a shortlist. It suits companies that already have real written material and verifiable third party references.
It does not suit a walk in local business. A tyre shop or a tiffin service is found by people searching nearby with intent to buy today and that demand is served by a map listing rather than an assistant answer. It does not suit a business with a thin website and no presence anywhere else because the groundwork is the website and paying for this first wastes the fee. It does not suit anyone who cannot currently tell you where their last twenty enquiries came from. Without that record there is no way to know whether this work ever paid and you will be judging it on screenshots.
What actually influences the answer
Consistency of facts comes first. If your company name category locations and the market you serve differ across your site directories and listings then a model has contradictory evidence and tends to hedge or omit. Correcting that is unglamorous and it is the highest value step available.
Next is being present in the sources these systems reach for. Your own pages trade publications review platforms recognised directories and question and answer communities all feed retrieval. Then write so a machine can lift a clean answer. State the question as a heading and answer it in one self contained paragraph immediately below naming the business and the specifics rather than referring back to earlier text.
Finally make sure you are readable. Server rendered content structured data and a sitemap all help. Check your robots file as well because every assistant publishes the crawler it uses and a site that blocks them cannot be retrieved. Paying for this work while blocking the crawlers is a contradiction worth checking before anything else.
How to tell whether it is working
Fix a list of twenty questions a real buyer would ask. Run them monthly across each assistant. Record exactly what came back including which sources were cited. Variation between runs is expected so look at the trend over several months rather than at any single answer. Alongside that keep asking every new enquiry where they heard of you and watch for assistant referrers in your analytics.
The flattering metric is a screenshot of one good answer. It proves nothing because the next run may differ. Visibility scores from tools that run their own undisclosed prompts flatter in the same way. They measure the tool rather than your market.
The mistake that wastes the budget
The common error is commissioning a hundred machine written question pages to feed the machines. It is expensive because the text is indistinguishable from what every rival produced the same month so it changes nothing in retrieval. It also puts the site at risk with ordinary search which is still where most of your demand is.
No citation can be promised and the reason is structural. The answer is generated afresh for each question by a system whose weights and retrieval behaviour change without notice or explanation. Influence over the inputs is real. Control over the output does not exist for anybody.
This guide explains how something works. It is not legal, tax or financial advice, and rules change. Confirm anything regulatory with a qualified professional before you act on it.