Can you guarantee ChatGPT will recommend my business?

Why no honest firm can guarantee it

Four things sit between any firm and that promise, and not one of them is inside its control.

• The output is not deterministic. A model writes each answer by sampling, so the same question can return different businesses on the same day from the same account. Nobody can guarantee the result of a draw they do not make. The mechanism is in why AI answers change every time you ask.
• There is no ranking lever. There is no slot, no bid, no placement and no position, because there is no ranked list to hold one. We are aware of no product from OpenAI or Google that sells placement inside a generated answer, and our own relationship with both is ordinary paid API access. The full version of that correction is how do I rank number 1 in ChatGPT.
• The material belongs to third parties. Answers are assembled from reviews, directories, publications and other people's pages. A firm can earn you a mention on somebody else's site. It cannot commit to one, because the decision is not its own, and it cannot commit to how an engine weighs it afterwards.
• The models change without notice. A model update, an index refresh or a change to how a provider grounds its answers can move your result when nothing in your business changed. No contract holds a third party's roadmap still.

Any one of those four would be enough on its own. A guarantee has to defeat all four at once.

What a firm can commit to instead

Three things, and you can check all three without having to trust the firm.

• Measurement you can reproduce. A pinned set of buyer questions, asked on a stated date, with every answer stored word for word and the mentions counted mechanically. Ours is 70% a count a third party can reproduce from the stored transcripts, and 30% model-graded recognition and sentiment, which they cannot. That split is published on the technology page rather than smoothed over.
• Defined work with a definition of done. Not improve your AI visibility. Named pages written, named facts corrected, named profiles claimed, named proof published, each one either delivered or not, agreed before the work starts.
• Evidence published either way. A re-scan on the same questions after the work, released whether the number moved or not. A firm that only shows you the re-scans that moved is running a screenshot business with extra steps.

There is a fourth commitment that matters more than the three, and it is a refusal: to tell you when the honest answer is that this work is not worth buying yet. Our version of that test, from a firm that sells the work, is on is AI visibility worth it.

A guarantee is a representation about a future matter

Under the Australian Consumer Law, a representation about a future matter is taken to be misleading if the person making it did not have reasonable grounds for it, and the burden of showing those grounds sits with them rather than with you.

We are not lawyers and this is not legal advice. It is a practical suggestion. If you are offered a guaranteed AI result, ask for the reasonable grounds in writing, in the same document as the guarantee. The request is polite and costs nothing, and a guarantee that cannot survive it was never meant to be enforced.

Our own refusal, with a date on it

We do not promise a score, and we published our own poor one rather than describe ourselves as an authority.

On 31 July 2026 we ran our own scanner over answerable.com.au. Site readiness came back four out of four. ChatGPT named us in none of the five buyer questions it was asked. The score was 29 out of 100. That is the honest picture of a well-built site with no external footprint, which is the exact failure this firm sells the fix for. The write-up is on the about page, alongside the three facts that page still cannot state, including that no founder is named on this site yet.

The limits of the measurement are published in the same spirit. Two engines rather than five. Scheduled rather than realtime. No access to anyone's real ChatGPT or Gemini sessions and no query volume data. No partnership, integration or authorised access with any AI provider. They are set out on the technology page, in the same size type as everything else.

Five questions to ask instead

Replace the guarantee question with five that have checkable answers, and read the hesitation as data.

1. Which exact questions will you ask, and will they be pinned so the next run is comparable to this one?
2. Will every answer be stored word for word, and can I read them in full rather than as a summary?
3. Is the count mechanical, or is a model asked to produce the number? If a model produces part of it, what share?
4. What is the definition of done for each deliverable, written down before the work starts?
5. What is my number today, and will you publish the re-scan on the same questions whether it moves or not?

A firm that can answer all five is selling measurement and work. A firm that answers with a guarantee is selling neither. What that work should cost in Australia, with real published prices and the red flags in a quote, is on what AI visibility costs in Australia.

One more promise to treat the same way: a dollar value per AI mention. Nobody has credibly measured it, us included, so a quoted figure is invented. How much a ChatGPT mention is worth gives you the method to build your own instead.