5 Prompt Tests to See If ChatGPT or Gemini Use Your Content

Prompt tests are a simple way to investigate whether ChatGPT or Gemini use your content when answering questions related to your business. You do not need API access, browser extensions, private account data, or a suspicious-looking spreadsheet with 47 tabs. You need a public page, a realistic customer question, and a repeatable record of what the AI says.

This distinction matters because AI visibility is not the same as Google ranking. A page can rank on page one and still never appear in an AI answer. Another page may receive a mention because it clearly explains a narrow question, even if it has modest traditional search traffic.

A useful test looks for more than your brand name. It checks whether the model recognizes your business, repeats a distinctive fact from your page, links to the page, or uses the page as support for a recommendation. Those are different levels of visibility and should be recorded separately.

The Google Search Central documentation on AI features explains that the same foundational practices that help pages appear in Google also support visibility in AI features. That does not mean every ranking page will be cited, but it does reinforce a sensible approach: publish useful, accessible information and measure what happens.

A citation log becomes much more useful when it includes timing. Start with the page publication date, the date it was first indexed, the date it first received an impression, and every date you ran a prompt test. Google Search Console can help you inspect impressions, clicks, queries, and pages through its Performance report.

Do not expect Search Console to show “ChatGPT citation” as a standard channel. Instead, use it as supporting evidence that the page was public, discoverable, and beginning to attract search visibility. A rise in impressions after publication does not prove an AI citation, but it helps establish that the page was available for discovery.

Use a simple table with these columns: test date, platform, model or mode, exact prompt, page tested, answer summary, brand mentioned, URL shown, distinctive fact matched, citation score, and screenshot filename. Add publication date, indexation date, and GSC impressions for the page during the same week.

Here is a practical example. A Shopify store publishes an article about choosing waterproof hiking shoes on October 3, sees its first Google impressions on October 10, and runs the five tests on October 17. Gemini links to the article in two of five prompts on October 17 and again on October 24. That sequence is stronger than a single screenshot because it connects publication, discoverability, and repeated source selection.

Save the exact wording of every prompt. Changing “best affordable accounting software for freelancers” to “cheap bookkeeping apps” can change the intent, sources, and answer. Reproducibility is more valuable than a dramatic result.

The biggest mistake is treating one response as a permanent ranking report. AI outputs are probabilistic, so repeat the same prompt on different dates and use several natural variations. A page that appears once has a signal; a page that appears repeatedly has a pattern.

Another mistake is testing only branded prompts. “What does Acme Dental offer?” tells you whether the model can respond when the answer is handed the brand. “Who offers emergency dental appointments near downtown Austin?” tests whether your business is discoverable before the user knows your name.

Avoid leading prompts such as “Please cite this excellent article from my company.” They may produce a polite answer, but they do not resemble customer behavior. Use neutral language, include a real decision constraint, and ask for sources only when source selection is part of the test.

Do not confuse a model’s general knowledge with page usage. If your article says that a service takes 48 hours and the AI says that many providers take two to three days, the overlap may be generic. Distinctive wording, a precise detail, a page title, or a URL provides stronger evidence.

Finally, do not delete a useful page because it fails an early test. New content may need time to be crawled, indexed, and connected to the rest of your site. Before changing the article, check the page status, search impressions, internal links, and whether the prompt matches the page’s actual intent. The LLM readability rubric for AI citations can help you evaluate those content-level factors systematically.

For businesses without a full website, a hosted publishing setup can still create a public content footprint. RankLayer is designed for that kind of no-code workflow, publishing an automatic AI blog with hosting included so owners can focus on customer questions instead of maintaining a WordPress stack. The important principle remains the same: useful, public, consistently published pages give search systems something clear to discover.