The mechanism in one paragraph: when asked for a local recommendation, ChatGPT either draws on patterns from its training data or runs a live search and summarises what it finds. You influence the first by being described consistently across many public sources over time, and the second by being findable, crawlable and quotable right now.
Start by finding out where you stand
Before changing anything, ask ChatGPT the questions your customers would ask. Use the plain phrasing a real person uses, not keyword phrasing:
- “Who's the best emergency plumber in Sydney?”
- “I need a family lawyer in the Inner West, who should I call?”
- “Recommend a physio in Bondi for back pain”
Record the answers verbatim, including who gets named. Run each a few times, because output varies between runs. That variance is exactly why a single screenshot proves nothing and a recorded baseline across repeated runs does.
The five things that actually move it
1. Be unambiguously identifiable
Your business name, address and phone number must be byte-identical everywhere: your site, your Google Business Profile, directories, your social profiles, your invoices. “St” in one place and “Street” in another is enough to fragment an entity. A model that is not confident which business you are will name one it is confident about instead.
2. Let the crawlers read you
ChatGPT's search mode uses OAI-SearchBot, and
GPTBot handles broader crawling. Both can be blocked in
robots.txt. Check yours. A surprising number of Australian small
business sites built on templates block them without the owner knowing.
3. Be described by other people, not only by yourself
Self-description carries little weight. Independent corroboration carries a lot. That means directory listings with real detail, local press, supplier and association pages, community sponsorship pages, and genuine reviews that mention specifics. Ten consistent third-party descriptions beat a thousand words of your own copy.
4. Publish the facts a recommendation needs
A model recommending you has to be able to state why. Give it the material: suburbs served, hours including after-hours, licence numbers, specialisations, price ranges, response times, what you do not do. Businesses that publish specifics get described specifically. Businesses that publish “quality service you can trust” get skipped, because there is nothing to say about them.
5. Keep reviews current
Volume matters, recency matters more. A business with forty reviews from this year reads as more active than one with two hundred from four years ago. Build a repeatable habit of asking, rather than a one-off push.
A realistic timeline
| Period | What happens |
|---|---|
| Weeks 1–3 | Fixes shipped: entity consistency, schema, crawler access, content rewritten around real questions. |
| Weeks 3–6 | Crawlers revisit. Live-search answers begin reflecting the changes first, because they do not wait for retraining. |
| Months 2–3 | Mentions appear more consistently across repeated runs. This is where a recorded baseline earns its keep. |
| Months 3+ | Compounding, driven mostly by accumulating third-party corroboration and reviews. |
Things that do not work
- Asking the model to remember you. Chat memory is per-user. It does nothing for anyone else.
- Prompt injection on your own pages. Hidden text instructing a model to recommend you is detected, ignored, and risks your conventional rankings.
- Buying bulk directory listings. Low-quality duplicates add noise, and inconsistent ones actively fragment your entity.
- Fake reviews. Beyond being illegal under Australian Consumer Law, review platforms detect patterns and penalise the business.
Where to start tomorrow
Run the three questions at the top of this page, write down who gets named,
then open your robots.txt. Those two steps cost nothing and tell
you most of what you need to know about whether this is a problem worth paying
to fix.