The Method, in One Paragraph
Write down twenty prompts a real buyer would type. Run each one across ChatGPT, Gemini, Perplexity and Google's AI Overviews. For every prompt, record two things separately: whether your business was named, and whether a page of yours was linked. Date it. Repeat monthly. That's the whole method, it takes about ninety minutes, and nothing else currently available does the job.
Why There's No Tool for This
Search Console tells you what Google showed you for. Nothing equivalent exists for AI answers, and the reason is structural rather than temporary.
AI systems are non-deterministic. Run the same prompt twice and you can get different businesses named, in a different order, citing different sources. There is no fixed ranking to report.
Commercial trackers do exist. SE Ranking, Semrush and others sample prompts at scale and report how often a brand appears. They are useful for benchmarking against competitors, and we use them. What they cannot tell you is whether your twenty buyer prompts name you, because those prompts aren't in their sample.
And two engines aren't tracked by anyone at all. No commercial tool currently reports Claude citations or Grok citations. Not SE Ranking, not Semrush, not the AI-visibility startups. If an agency tells you they track your Grok visibility, ask to see the raw output.
Writing Prompts That Are Worth Measuring
This is where most attempts go wrong, and it's the only part requiring judgement.
The instinct is to write prompts containing the phrase you want to rank for. Best AI marketing agency Melbourne. Nobody types that into ChatGPT except people in marketing.
What people actually type is closer to how they'd ask a knowledgeable friend. A Sagapixel study reported by Search Engine Land found 75% of ChatGPT sessions about local services used keyword-style searches — short, like a Google query — but the ones that produce recommendations are longer and situational.
Write prompts in four shapes, roughly five each.
Situational. I run a plumbing business in Brunswick and I'm spending $2,000 a month on Google Ads with no idea if it's working. Who should I talk to?
Direct recommendation. Recommend a Melbourne agency for Facebook and Instagram ads for a small retailer.
Comparison. What's the difference between an AI marketing agency and a normal digital agency, and is the difference worth paying for?
Category. Which Australian agencies specialise in getting businesses cited by ChatGPT?
Then one rule that improves the whole exercise: write the prompts before you look at the answers. Writing prompts you already suspect you'll win is how this becomes theatre.
Scoring: Two Columns, Not One
Keep brand mention and link citation separate, because they have different causes and different fixes.
Brand mention means the model named your business in the text. This comes mostly from sources you don't control: directories, listicles, review platforms, press, forum threads.
Link citation means the answer linked one of your pages. This comes from your own content being crawlable and extractable.
The distinction is visible in real data. In our October 2026 pull of the Australian market, Impressive Digital recorded 572 brand mentions and 8 link citations. Online Marketing Gurus recorded 1,076 and 3,050. Same category, opposite profiles, and completely different work required to improve each.
A simple scoring sheet, one row per prompt:
| Column | What goes in it |
|---|---|
| Prompt | The exact text, unchanged between months |
| Engine | ChatGPT, Gemini, Perplexity, AI Overviews |
| Named | Yes / No |
| Linked | Yes / No, plus which URL |
| Who else | Every competitor named |
| Date | The date you ran it |
That last-but-one column is the one people skip and it's the most useful. Knowing which three competitors keep appearing tells you where the citations are coming from, and you can go and look.
Running It Consistently
Four things that make the numbers comparable month to month.
Same prompts, word for word. Changing a prompt resets the series. Keep a locked list.
Fresh session each time. Previous conversation affects the answer. New chat per prompt.
Location set. For local queries, set your region explicitly where the tool allows it, and note it. An Australian query answered against a US default is a different question.
Same week of the month. Model updates land continuously. Running in the first week of each month keeps the comparison roughly honest.
Log it in a spreadsheet, not a document. You want to count.
What We Found When We Ran It on Ourselves
We publish this because the alternative is asking you to trust a method we haven't shown working.
As of October 2026, across all five engines that can be tracked commercially — AI Overviews, Google AI Mode, ChatGPT, Gemini, Perplexity — Inflo Media records zero brand presence and zero link presence.
Twenty-plus published articles. A full service-page build across fourteen pages. Nothing yet.
That isn't a comfortable number to put on our own website. It's also the only thing that will make November's figure mean anything, and it's a fair illustration of the timeline: AI citation moves on a three-to-six-month horizon, and we're inside that window rather than past it.
What it also demonstrates is the thing we tell most businesses who ask about AI visibility. Publishing is necessary and nowhere near sufficient. The citations follow from third-party corroboration, entity consistency and crawler access, and those take longer than writing does. Why businesses don't show up in AI search covers the sequence.
What to Do With the Results
If you're not named and not linked, the problem is foundational. Thin site, few reviews, no third-party mentions. Fix those before anything else.
If you're named but not linked, your reputation precedes your content. Make pages extractable: direct answers under headings, FAQ sections, clean structure. How to appear in Google AI Overviews has the specifics.
If you're linked but not named, your content is working and your entity isn't. Make business details identical everywhere and connect them with sameAs structured data.
If competitors keep appearing and you don't, go and read why. Open the sources the answer cites. Usually it's a directory you're not on or a listicle you're not in.
If you'd rather not run it yourself, that's our free AI visibility audit: twenty prompts, four engines, dated, and you keep the results whether or not you work with us. Our AI marketing page covers the service side.
Frequently Asked Questions
How do I know if ChatGPT recommends my business?
Run twenty buyer prompts in fresh ChatGPT sessions and record whether your business is named and whether a page of yours is linked, keeping the two as separate columns. Date the results and repeat monthly with the identical prompts. There is no report or dashboard for this. Manual sampling is currently the only reliable method.
Is there a tool that tracks AI search visibility?
Several exist and they sample prompts at scale, which makes them useful for competitor benchmarking. What they can't tell you is whether your specific buyer prompts name you, since those aren't in their sample. No commercial tool currently tracks Claude or Grok citations at all, so treat any claim to do so with scepticism.
What's the difference between brand mention and link citation?
A brand mention is the model naming your business in its answer; a link citation is the answer linking one of your pages. They come from different places: mentions from third-party sources like directories and listicles, citations from your own content being crawlable and extractable, so they need different fixes and should be scored separately.
How many prompts should an AI visibility audit use?
Fifteen to twenty is the practical range. Fewer and single-answer variation swamps the signal; more and you won't sustain it monthly. Spread them across four shapes: situational scenarios, direct recommendation requests, comparisons, and category questions.
How often should I check my AI visibility?
Monthly, in the same week each time. Model updates land continuously, so a fixed point in the cycle keeps month-to-month comparison roughly honest. Checking weekly produces noise; checking quarterly means you won't know what caused a change.
Why do AI tools give different answers to the same question?
They're non-deterministic by design. The same prompt run twice can name different businesses in a different order citing different sources. This is why single checks are meaningless and why any guarantee of a specific position in an AI answer is a claim about something nobody controls.
What does it mean if no AI tool mentions my business?
Usually that there isn't enough about you for a model to assemble an answer from. AI systems synthesise from directories, review platforms, press and your own site. A thin website, few reviews and no mentions you don't control gives them nothing to work with. The fix is foundational rather than clever.
Can I track my visibility in Claude or Grok?
Only manually. No commercial visibility tool reports citations from either, so the method is to run your prompts in each and record the results by hand. For Grok specifically, be aware that its distinguishing feature is a native connection to X's live stream, so presence on X affects results in a way that website work does not.
How long before AI visibility work shows results?
Three to six months is the realistic window for ChatGPT, which refreshes its understanding of a business slowly. Google's AI Overviews can move faster because they draw on the live search index, sometimes within weeks of a strong page. Nobody controls the timing, and a dated baseline is what lets you tell progress from noise.
Should I pay an agency to track AI visibility?
Only if they'll show you the raw results rather than a summary score. The method is laborious but not complicated, and an agency genuinely doing it will hand over the prompt list, the dates and the unedited answers. A dashboard with a single visibility number and no underlying data is a product, not a measurement.