ChatGPT Recommends Your Competitors. Now What?
ChatGPT recommends competitors and not you. How AI assistants actually pick local businesses, which levers are real, and how to audit your own visibility.
Type "best [your service] in [your city]" into ChatGPT and it will hand back a confident list of five or six businesses, each with a tidy sentence explaining why they're good. If yours isn't on that list, the natural reading is that you've been assessed against your competitors and come last. That reading is wrong, and acting on it is how business owners end up on a retainer for something nobody can deliver.
What actually happened is more mundane. The assistant ran a web search, pulled back a handful of pages, and summarised them. It did not evaluate your business. In most cases it has never encountered your business. The list you're staring at is a compression of two or three articles someone else wrote about your industry, plus whatever directory happened to rank that day.
Which gives you a much more useful question than "how do I rank in ChatGPT". The question is: which pages did it read, and why isn't your name on them? That's answerable in twenty minutes, and the fixes are ordinary work — the sort of work that also happens to move your position in normal search. We covered the mechanics of AI answers generally in how to show up when customers ask AI instead of Google. This is the narrower, more panicked version: you looked, you weren't there, and you want to know what's real.
What the assistant is doing when it names five businesses
Three separate things feed a local recommendation, and they behave very differently.
What the model remembers. Models are trained on a snapshot of the web that ends at a fixed point. For a nine-person accounting firm in Ballarat there is almost certainly nothing memorised — small Australian businesses don't survive the compression of training. Very large brands do. This is the one input you cannot influence at all.
What it searches right now. Any decent assistant answering a "best X in Y" question goes and searches. That grounding step is where the actual names come from, and it means the answer is downstream of a live search results page — with all the same winners on it. The businesses being named are, overwhelmingly, the businesses already visible for that query in ordinary search.
What the local data layer contains. Directories, "top ten" listicles and comparison pages are themselves largely assembled from Google Business Profile listings, categories and review counts. So your Business Profile data reaches the answer even when the assistant never touches a map — it arrives second-hand, baked into a page some publisher generated eighteen months ago and hasn't touched since.
Stack those three together and the picture is unglamorous: AI recommendations for local services are mostly a re-run of who was already winning, laundered through a friendly paragraph. The upside is that you already know how to compete there.
Ask it three times and you get three answers
Before you spend a dollar on this, understand how weak the evidence is that set you off.
Run the same query in a fresh session and the list changes. Run it logged out and it changes again, because your account's memory was quietly shaping the earlier answers. Swap "best" for "most reputable" and half the names swap with it. None of that is a ranking system. It's a summarisation step with sampling built in, sitting on a search index that reshuffles daily.
So one absent result is not a diagnosis. Five absent results across five phrasings, in a clean session, with the same three competitors every time — that's a diagnosis. Test properly before you react.
The citations underneath the answer are the whole map
This is the part most people scroll past. When an assistant answers a grounded question, it shows its sources — small numbered links, or a "sources" strip under the response. Click every one.
What you'll find is remarkably consistent for Australian service queries: a couple of "best [service] in [city]" listicles from local publishers or lead-generation sites, one or two directories, an industry association's member search, occasionally a Reddit thread, and sometimes one or two actual business websites — usually the ones with a page targeting that exact phrase.
That set of pages is your target list. Not ChatGPT. You cannot optimise a model, but you can absolutely get your business added to a directory, listed with your professional association, and named in the local publication that keeps writing the roundup. Whether those particular pages deserve the authority they've been handed is a separate argument, and one the model is not having.
Two things follow. Third-party presence is now a visibility asset rather than a vanity one. And your own site still has to close the loop — the moment someone hears your name from an assistant, their next move is to search it, and if your site isn't the clean first result for that name, the recommendation evaporates. That's ordinary small-business website work, not AI strategy.
What you can genuinely move
One caveat on reviews that catches out a whole category of business. If you're a regulated health service — physiotherapy, psychology, optometry, dentistry, most cosmetic medicine — AHPRA's advertising rules prohibit using patient testimonials in advertising, and that includes reviews you solicit and display. The review lever is genuinely narrower for you. What replaces it is credentials, clear clinical information and referral relationships, and the fact that most of your competitors are quietly breaching the rule is not a reason to join them.
For everyone else: buying reviews is illegal under Australian Consumer Law and the ACCC has taken action over fake and misleading ones. The shortcut is not available. We've written up the version that works in a Google reviews strategy that actually feeds your website.
What nobody can sell you
Guaranteed placement in AI answers. There is no inclusion mechanism to buy — Google's own documentation states that no special file, markup or structured data is needed to appear in its AI features, and the other assistants publish exclusion controls, not inclusion ones. A guarantee here is a guarantee about something that does not exist.
"AI rank tracking" dashboards. Given that the same query returns different names in different sessions, a chart plotting your position over time is largely plotting sampling noise with a trend line drawn through it. Worth watching directionally at most; not worth a monthly fee.
Prompt-stuffing pages — writing content built around imagined ChatGPT queries. Assistants aren't string-matching prompts, and the pages read badly to the humans who eventually land on them.
Paid inclusion in "AI citation networks", which is a link scheme wearing a new lanyard.
And schema markup positioned as the AI lever. Structured data is worth doing properly — it removes ambiguity about your address, hours and prices — but pretending it's what stands between you and a recommendation sells an afternoon's work as a strategy.
A 20-minute audit you can run yourself
- Open a logged-out session in a private window. No account, no memory, no history bleeding into the result.
- Ask five phrasings of your core query — "best", "most experienced", "who should I use for", "recommend a", and one with a specific qualifier a real customer would use ("open Saturdays", "does commercial work").
- Record who appears. If the same three competitors show up in four of five answers, that's the real competitive set.
- Click every citation and list the source pages. This is the deliverable — the actual map of what's feeding the answer in your category.
- Check each source page for your business. Missing from a directory? Add yourself. Missing from a listicle? Note the publisher and pitch them.
- Ask the assistant about your business by name. If it can't describe what you do accurately from your own site, the problem is your copy, not the model.
- Read your robots.txt for blanket disallows against AI search crawlers.
- Search your own business name in ordinary Google and confirm your site is the first result with a sensible description under it.
That's the whole job. It costs nothing but attention, and it will tell you within half an hour whether you have a visibility problem, a website problem, or a bad afternoon.
FAQ
Why does ChatGPT recommend my competitors and not me?
Almost always because your competitors appear on the third-party pages the assistant searched — directories, association listings, "best in city" roundups — and you don't. The model isn't comparing your businesses. It's summarising pages, and your name isn't in them. Find the citations under the answer and you'll usually see the gap immediately.
Can I pay to appear in ChatGPT or Gemini recommendations?
No. There is no paid inclusion product for AI recommendations from any major assistant, and Google states plainly that no special markup or file gets a page into its AI features. What you can pay for is the underlying work — directory presence, review generation, content and technical fixes — which is ordinary marketing under a newer name.
How long does it take to start appearing in AI answers?
There's no honest fixed number, but the components have different clocks. Directory and association listings can land within days to weeks. Review volume builds over months. Your own site's search visibility moves on normal SEO timescales, which is months rather than weeks. Anyone quoting you a specific date is guessing.
Is AEO a real thing or just SEO with a new name?
Mostly the second, with a small genuine addition. The real, distinct work is checking crawler access, making sure your facts exist as plain sentences on pages you own, and getting into the third-party sources assistants cite. Everything else on a typical "AEO" invoice is technical SEO and content with the headings renamed.
Does blocking AI crawlers hurt my business?
It depends which one. Blocking a training crawler is a defensible commercial decision and doesn't remove you from search-grounded answers. Blocking a search crawler removes you from that assistant's answers entirely. The two are separate user agents, and many sites blocked both years ago without realising there was a difference.
Where to point the effort
The uncomfortable, useful conclusion is that there is no separate AI channel to win. There's your business, described consistently, present on the pages that describe your category, with enough recent reviews to look alive, on a site fast and clear enough that a machine can read it and a human can act on it. Do that and you show up in AI answers as a side effect of showing up generally — the same way the competitors outranking you in ordinary search rarely got there with one clever trick they bought.
What's genuinely changed is the cost of being invisible. When the answer replaces the click, absence from the answer is absence from the market for that query. We've written about what that does to a traffic graph while your rankings sit perfectly still, and it's the reason this is worth twenty minutes of your attention rather than none.
Start with the audit above. If it turns up a site problem rather than a listings problem — pages that don't load, copy that never says what you do, a structure nothing can parse — run your URL through our free audit for the technical half of the picture, or book 20 minutes and bring the source list you collected. Custom builds start at $8,000 AUD, and half the time the conversation ends with us telling you the build isn't the problem. Either way you'll leave knowing which one it is.