Test AI visibility with a repeatable prompt set, not one branded search. Run the same service, location, comparison and trust prompts in ChatGPT, Gemini, Perplexity and Google AI. Record whether your business appears, whether its services and location are accurate, which sources support the answer and whether the result gives a usable route to your website or enquiry page.
A branded prompt such as “Tell me about IDJoy” proves very little.
The engine already has the business name. A buyer often starts somewhere else: “website designer in Pretoria”, “who can improve a service-business website that gets traffic but no leads?”, or “compare three South African web designers for a small company”.
Those prompts test discovery. Discovery is the commercial problem.
Use twenty prompts that reflect the questions a buyer could ask before they know your name. Run the same set across ChatGPT, Gemini, Perplexity and Google AI features. Keep the results so the next test has a baseline.
Test four kinds of buyer intent
A useful baseline covers more than category keywords.
Start with four groups.
Service prompts
These ask for the work you sell.
Examples:
- website designer for a South African service business;
- brand identity designer in Pretoria;
- company that can fix a website with poor conversions;
- web designer that can connect a website to WhatsApp enquiries.
Write the prompts in the language your customers use. If customers say “website company”, include that phrasing even if your preferred term is “web design studio”.
Location prompts
Local prompts test whether the engine understands where you work.
Use the real operating model. A business based in Pretoria that serves Johannesburg should test both cities. It should not claim a Johannesburg office if one does not exist.
Examples:
- web designer in Pretoria;
- website company serving Johannesburg businesses;
- South African web designer for a Gauteng engineering company.
Comparison prompts
Comparison prompts expose the shortlist.
Ask the engine to compare three or five providers. Add a real decision factor such as price transparency, local experience, case studies, turnaround time or a particular service.
Record every competitor that appears repeatedly. Their websites and third-party footprint become useful research material.
Trust prompts
A buyer may ask for a trusted, experienced or well-reviewed provider rather than a service alone.
Test prompts such as:
- trusted website designer in Pretoria;
- South African web design company with case studies;
- website designer with clear pricing and business contact details.
These prompts show whether the engine can find evidence that supports the trust claim.
Record more than a yes or no
A mention can still harm the business when the answer uses an old phone number, wrong location or service you stopped offering.
Score each result on five fields.
| Field | Record |
|---|---|
| Mention | Did the engine name the business? |
| Accuracy | Were the service, location and business facts correct? |
| Source | Which website, profile or third-party page supported the answer? |
| Commercial fit | Did the prompt match work you want to sell? |
| Next action | Could the user reach a useful website or enquiry route? |
Do not combine those fields into a single vanity score too early. A business that appears in ten prompts with inaccurate information has a different problem from a business that appears in only two prompts with accurate information.
Run the prompts under repeatable conditions
AI answers change. Search results also change by location, product surface, account context and query wording.
Keep the test controlled enough to compare one month with the next.
Record:
- date;
- engine or product;
- exact prompt;
- whether you were signed in;
- location context if it affected the query;
- business mention;
- cited or linked sources;
- screenshot or copied answer for internal reference.
Do not rerun a prompt five times and choose the answer you like. Record the first clean result, then use a second run only when you are checking whether the answer is stable.
Read each engine on its own terms
ChatGPT Search can search the web and link to sources. OpenAI says sites should allow OAI-SearchBot if they want content available for summaries and snippets in ChatGPT search. OpenAI also states that no site can guarantee top placement.
Perplexity describes itself as an AI-powered search engine that searches the web and provides cited answers. Its source mix can differ from ChatGPT even when the prompt stays the same.
Google says the normal Search foundations remain relevant for AI Overviews and AI Mode. A page must be indexed and eligible to appear in Google Search with a snippet before it can qualify as a supporting link. Google does not require a special AI schema or AI text file.
Gemini can use Google services and web information, but the exact answer can still differ from Google Search. Treat Gemini as its own test surface.
The differences matter because “AI visibility” is not one ranking position.
Diagnose the reason before changing the website
Group failed prompts into four problems.
The engine cannot access the useful page
Check robots.txt, CDN rules, status codes, canonical tags and whether the important content exists in readable HTML.
Technical access comes first. Rewriting the page will not help a crawler that cannot retrieve it.
The engine cannot identify the business clearly
Compare the business name, services, phone number, location, website URL and business description across the site, Google Business Profile and major third-party profiles.
Fix contradictions at the source.
The website lacks evidence
A service page that says “we deliver world-class solutions” gives an engine little to verify.
Publish specific services, process, locations served, project evidence, case studies, pricing context and author or company information where relevant. Link claims to primary sources when you quote standards, laws or statistics.
Other sources describe competitors more convincingly
AI search can use third-party sources. Reviews, directories, association pages, client mentions and editorial coverage can strengthen or weaken the picture of a business.
Do not manufacture mentions. Build the public footprint through real work, accurate listings, legitimate partnerships and customer feedback.
Connect visibility to enquiries
A mention has limited value if the cited page has no useful next step.
Check the page the engine sends people to. A buyer should be able to understand the service, confirm that the business serves their area and choose a contact route without searching the rest of the site.
Track AI referral traffic where the platform exposes it. OpenAI says publishers that allow OAI-SearchBot can track referral traffic from ChatGPT in analytics. Google introduced dedicated generative-AI performance reporting in Search Console for a subset of sites in June 2026, while its AI-feature traffic remains part of overall Search performance reporting.
Pair traffic with business events:
- form submissions;
- WhatsApp starts;
- phone clicks;
- qualified enquiries;
- proposals;
- sales where attribution is credible.
A traffic increase without a usable enquiry route leaves the commercial problem unsolved.
Repeat the baseline after meaningful changes
Run the same twenty prompts after you change business information, publish important service pages, add verified case studies or correct crawler access.
A monthly test is frequent enough for many small businesses. Weekly testing can create noise unless the business publishes or changes information at that pace.
Keep the old results. The value sits in the comparison: which prompts improved, which engines changed, which incorrect facts disappeared and which competitors continue to dominate the shortlist.
IDJoy uses this baseline as part of its AI search visibility work. The service focuses on access, business clarity, evidence, third-party consistency and measurable enquiries rather than guaranteed AI rankings.