Most audits of this kind begin with the website and end with a long list. Ours begins with the assistants, because the only question that matters to a Sydney business is whether ChatGPT, Gemini, Perplexity or an AI Overview names it when a customer asks. Here is the afternoon, hour by hour.
Before you start: five pages and twenty Sydney prompts
Pick the pages that earn you money and the questions your customers actually ask, suburbs included.
Choose five pages: the homepage, your two most valuable service pages, the pricing page if you have one, and one area page. These are the pages you want quoted.
Then write twenty prompts the way a Sydney customer would type them into an assistant. Generic prompts tell you nothing; the suburb and the situation are what make them real.
| Business | Prompts worth testing |
|---|---|
| Physio, Inner West | “Who is a good physio in Marrickville for a running injury?” “NDIS physio near Newtown” |
| Family lawyer, Parramatta | “Best family lawyer in Parramatta for a property settlement” “How much does a divorce lawyer cost in western Sydney?” |
| Plumber, Northern Beaches | “Emergency plumber Dee Why open now” “Hot water system replacement cost Northern Beaches” |
| Accountant, Chatswood | “Small business accountant Chatswood” “Accountant for a café on the North Shore” |
Put the prompts in a spreadsheet with a column per assistant. You will fill it in during the first hour and again every month.
First hour: ask the assistants who they name
Run every prompt in every assistant and write down the businesses named and the pages cited.
Work through the twenty prompts in ChatGPT, Gemini, Perplexity and Google’s AI Mode, and note any AI Overview that appears on the equivalent Google search. For each answer, record which Sydney businesses are named and which URLs are cited or linked.
Two patterns show up quickly. Some prompts return the same three or four businesses across every assistant; those businesses have done something right that you can find in the next hour. Other prompts return directories and review sites rather than businesses, which tells you the assistants have not found a business page worth quoting.
Count how many of the twenty name you at all. For most Sydney businesses doing this for the first time, the answer is few or none, and that is the baseline you are going to improve.
Second hour: find out why the cited pages won
Open every page the assistants cited and compare it with your own, point by point.
For each cited page, note what it has in its first screen that yours does not. In our experience the list is short and repetitive:
- The opening sentences answer the question directly, before any introduction about the business.
- The page carries specifics an assistant can quote: prices or ranges, the suburbs served, turnaround times, a named person with a role.
- Reviews on the Google Business Profile mention the service and the suburb in the prompt.
- The page is plain HTML that loads without scripts, and the site does not block AI crawlers.
Now check your own five pages against the same list. For the last point, open robots.txt and any firewall or security plugin settings and look for GPTBot, ClaudeBot, PerplexityBot and Google-Extended. A block put in by a developer two years ago is a common and complete explanation for being invisible.
Then view the source of each page with scripts disabled. If the text is missing, several assistants are seeing an empty page.
Third hour: the fixes you can make today
Four changes fit in the last hour and cover most of the gap for a typical Sydney site.
- Unblock the crawlers you want to be cited by, in robots.txt and in the firewall, and confirm with a fetch from the server log the next day.
- Rewrite the opening of each of the five pages so the first two sentences answer the question in the page’s title, with the suburb or area named.
- Add one specific thing to each page that exists nowhere else: a price range, a turnaround, a photo of real work, the name of the person who does it.
- Update the Google Business Profile: primary category, services, hours, recent photos, and a reply to the last ten reviews that names the service.
Structured data belongs on the list too, but only when it mirrors the page: an Organization, LocalBusiness or Service block that claims reviews or prices the page does not show hurts trust rather than helping it. The schema guide covers what to include.
Before you leave: set up the measurement
Without a referral view and a saved prompt list, next month’s rerun has nothing to compare against.
- In GA4, build a view or exploration that isolates sessions from the assistant domains (chatgpt.com, perplexity.ai, gemini.google.com, copilot.microsoft.com), so AI referrals stop hiding in “referral”.
- Save the prompt spreadsheet with today’s results as the baseline, and put a monthly reminder in the calendar to rerun it.
- In Search Console, bookmark the informational queries where impressions are rising and clicks are not; that pattern usually means an AI Overview is answering above you.
- Ask whoever hosts the site for a monthly count of requests from the AI user agents.
This is the part most audits skip and the part that makes the rest worth doing. If nothing is measured, every fix is a guess.
Scorecard: how a typical Sydney site comes out
Score each area pass, partial or fail; the pattern below is what we see most often on a first run.
| Area | What a pass looks like | Typical first result |
|---|---|---|
| Named in answers | Named for most suburb-specific prompts | Named for few or none |
| Crawler access | No AI crawlers blocked | Partial: a security plugin blocking some |
| Rendering | Text present without scripts | Pass on most builders; fail on some app-style sites |
| Opening answers | First two sentences answer the title | Fail: pages open with the company story |
| Specifics | Prices, suburbs, people, real photos | Partial: suburbs yes, prices no |
| Profile and reviews | Current profile, reviews naming services | Partial: profile current, reviews generic |
| Measurement | Referral view, prompt list, log access | Fail: none of the three |
Fix in this order: access, measurement, openings, specifics, profile. A site that fails access should do nothing else until that is solved; a site that passes everything but has no specifics has a writing job, not a technical one.
If you would rather we ran the afternoon, it is part of our SEO audit, and the fixes are what our AI SEO service does month to month.
