Straight answer
Is AI visibility worth it?
The honest version, from a firm that sells it. What the evidence supports, what nobody can guarantee, and the cases where you should not buy this yet.
We sell AI visibility work, so read this with that in mind. It is also why we published our own scan result, which is bad, and the raw data behind every number below, which anyone can recompute.
The short answer
AI visibility is real but narrow. The measurable part is whether an assistant names your business when a buyer asks a question in your category. On 30 July 2026 we asked ChatGPT and Gemini about 300 Australian businesses: 41% of the 137 with complete records from both engines were never named once. That gap is real and it is fixable. Nobody can guarantee the outcome.
Why the scepticism is fair
Google's John Mueller argued publicly in August 2025 that the harder a new acronym is pushed, the more likely the person pushing it is producing spam or running a scam. He was talking about GEO, AEO and AIO. That is a reasonable prior, and the behaviour that earned it is easy to list.
- Guarantees of a position inside an AI answer, which no vendor controls.
- Scores with no published formula, so the number cannot be checked or argued with.
- Statistics with no sample size, no date and no link to a source.
- Retainers sold before anyone has measured whether there is a problem.
- Rebadged link building with an AI label on the invoice.
None of that means the underlying shift is fake. It means the category attracted sellers faster than it attracted evidence. The honest test is not whether the acronym is new. It is whether the specific claim in front of you comes with a number, a denominator and a date.
What our own data actually shows
Every figure below comes from one scan: 300 search-visible Australian businesses across six industries, put to ChatGPT and Gemini on 30 July 2026, with automated checks on all 300 websites. The raw rows are published under CC BY 4.0 so the arithmetic can be redone without asking us. Where a denominator is 137, that is the number of businesses with complete records from both engines.
| Finding | Figure | Counted over |
|---|---|---|
| Never named once by either engine | 41% | 137 businesses with complete two-engine records |
| Recognised when asked about directly, never named when a buyer asks who to use | 26% | The same 137 businesses |
| Average AI visibility score | 38.3 | All 300 businesses, median 36 |
| Block AI crawlers outright in robots.txt | 13% | All 300 websites |
| No JSON-LD structured data on the homepage | 24% | All 300 websites |
| Average on-site AI readiness score | 74.9 | All 300 websites |
| Correlation between site readiness and per-engine visibility | r = 0.116 | 137 complete two-engine records |
Read the last two rows together, because they are the most useful thing in the dataset and they cut against the people selling this. Australian business websites are, on average, in reasonable technical shape for AI. Being in good technical shape barely predicts whether an engine names you. A correlation of 0.116 is close to nothing.
That single number is the strongest argument against buying a package of on-page fixes as though it were the whole job, and it is our own data saying it. The full method and the per-industry splits are in the study, and the headline figures are collected on the statistics page.
What nobody can guarantee
No vendor, including us, can guarantee that an AI assistant will name your business. The answer is generated fresh each time, the models change without notice, and the same question can return different businesses on the same day. Anyone offering a guaranteed placement in an AI answer is either misunderstanding the system or misrepresenting it. What can be promised is measurement, published method and specific work.
There are three further limits worth stating before you spend anything.
- Attribution is genuinely hard. If an assistant shortlists you and the buyer then searches your name directly, no analytics tool records the assistant as the source. Part of the value of this work leaves no trail.
- Answers move. A result you earn can move next month. That is an argument for monitoring, not an argument that the work does not hold.
- Scores are estimates of a moving thing. Ours is 70% a count of how often you were named across the questions we asked, which a third party can reproduce from the stored answers, and 30% model-graded recognition and sentiment, which they cannot. We publish that split on the technology page rather than presenting the number as fully mechanical.
When it is worth doing, and when it is not
| Your situation | Our honest read |
|---|---|
| Buyers in your category already ask AI who to use, and you are never named | Worth doing. This is the case the evidence supports most directly. |
| You are recognised when asked about by name, but never recommended | Worth doing, and usually the cheapest fix. The engine knows you exist and has no reason to put you forward. |
| AI states your prices, hours or services wrongly | Worth doing. Correcting the facts an engine repeats is concrete work with a checkable result. |
| Your robots.txt blocks AI crawlers | Fix it yourself today. It is one file. You do not need an agency for this. |
| Your website is broken, slow or has no clear service pages | Not yet. Fix the site first. Nothing downstream compensates for a site that cannot answer the question. |
| You have no repeatable way to serve more customers | Not yet. Demand you cannot serve is not an improvement. |
| Someone is selling you a guaranteed AI ranking | Walk away. The thing being sold does not exist. |
The test we apply to ourselves
On 31 July 2026 we ran our own scanner over answerable.com.au. It returned 29 out of 100. All four site checks passed, and ChatGPT named us in 0 of 5 buyer questions in our own category.
We are publishing that because it is the clearest available demonstration of the thing we just said. Our on-page work is as good as it gets and it did not make an assistant name us, because almost nothing about us exists anywhere except our own website. A perfect site with no external corroboration scores badly, and it should. If a vendor cannot show you their own number, ask why.
How to decide in about ten minutes
- Open ChatGPT and Gemini. Ask each one the question a customer would ask, in your category and your city, without your business name in it.
- Write down who gets named. Repeat the question twice more; the answers will not be identical.
- Now ask each engine about your business by name. If it knows you but never volunteers you, you have a corroboration problem, not a website problem.
- Check whether your robots.txt blocks GPTBot, ClaudeBot, PerplexityBot or Google-Extended. If it does, that is your first fix and it is free.
- If you are named in a reasonable share of those answers already, spend your money elsewhere this quarter.
If you would rather not do it by hand, our free scan runs the same idea and shows you one real quote of what an engine said about you today. It costs nothing and there is no email gate, and the counting method is published in full.
Questions
Common questions from sceptics
Is GEO a scam?
Can anyone guarantee that AI will recommend my business?
Is AI visibility just SEO with a new name?
Does AI visibility work actually send anyone to my site?
What is the honest evidence that this matters?
How does Answerable score on its own scan?
When should I not buy AI visibility services?
Keep reading
Related pages
Diagnostic
Why doesn't ChatGPT mention my business?
Six causes, ranked by how often we found each across 300 Australian businesses.
Read more →Commercial
What AI visibility costs in Australia
The four price bands, real published tool prices, and what a fair quote includes.
Read more →Research
The State of AI Visibility 2026
300 Australian businesses scanned across ChatGPT and Gemini, raw data published.
Read more →Check the claim yourself.
The free scan runs the same questions we used in the study, on your business, and shows one real quote of what an engine said about you today. No email gate.