You do not need a monitoring subscription to learn whether AI assistants mention your brand. You need ten consistent prompts, fifteen quiet minutes and a sheet you can reuse. Aim for a small check you can repeat and compare month to month.
Run the same prompts in ChatGPT, Perplexity, Gemini and Google AI Mode. Use a fresh conversation each time. Keep your location and wording fixed. You are checking whether your brand appears, how it is described and which pages support the answer.
Run these ten prompts without coaching the answer
Do not put your brand name in the first eight prompts. If you do, you are testing recall after a hint. Replace the bracketed terms with your category, customer and location, then copy each answer into your sheet.
- What are the best [category] providers for [customer] in Singapore?
- Which [category] companies should a small marketing team compare?
- Recommend three providers for [specific problem] in Singapore.
- Who is known for helping [customer] with [outcome]?
- What are the alternatives to [well-known competitor] for a smaller team?
- Which provider is a good fit if [important buying constraint]?
- What should I look for when choosing a [category] provider?
- Compare the leading [category] options for [use case].
- What does [your brand] do, and who is it for?
- Is [your brand] a credible option for [specific use case]? Show your sources.
Use the assistants' normal free access where available. Access and limits change, so this audit measures the experience a buyer can reach on the day you run it.
Record the evidence in one sheet
Create one row for each prompt and assistant. Add the date, exact prompt, whether your brand appeared, its position in the answer, wording used, cited page, cited domain and any obvious error. Save the full answer or a screenshot as evidence.
Use three plain outcomes: mentioned, absent or prompted only. A mention is stronger when it appears before you name the brand. A prompted-only result still tells you whether the assistant understands the company once given a cue. Keep incorrect descriptions in the sheet. Wrong wording is a work item, and hiding it helps nobody.
Check how each assistant presents sources
The interfaces do not present evidence in the same way. OpenAI says ChatGPT search responses may include inline citations; a Sources control can show cited and related links when available. Perplexity says every answer includes citations that link to original sources.
Gemini may show a Sources button or inline links, but Google says not every response includes them. Google AI Mode places links within responses and around the answer. Google's August 2026 update for website owners says it increased inline links and added website previews.
A citation is not proof that the linked page supports the sentence beside it. Open the page. Check the wording. Record mismatches separately.
Also note whether the answer links to your homepage, a useful service page or somebody else's summary of you. Those outcomes need different fixes. A homepage mention suggests the assistant knows the company exists. A specific page citation tells you which piece of evidence it found useful. And if a third party is doing the describing, somebody else is writing your public story.
When your brand is absent, fix the evidence
Do not respond by publishing twenty vague articles. Start with the missing buyer question. Give that question a clear page with a direct answer, concrete scope, author, date and useful proof. Make your company name, category, location and customer fit consistent across your site and public profiles.
Then inspect what the assistant did cite. If it favours directories, comparison pages or original guides, ask whether your brand belongs there on merit. If competitors explain the category more clearly, improve your answer. We think AI visibility usually exposes an evidence problem long before it exposes a prompting problem.
Repeat often enough to spot movement
Run the check monthly while you are actively improving pages. Quarterly is enough once results settle. Keep the same eight discovery prompts and two branded prompts for at least three rounds. Add a new prompt only when buyers begin asking a materially different question.
Your next step is to open a blank sheet, add forty rows and run prompt one across all four assistants. Fifteen minutes later you will have a baseline. Next month you will have something to compare it against.