The exact question we used

Which dental practices in South Austin, Texas offer dental implants, and what should I check before choosing one? Use current web sources.

On September 16, 2026, we entered this question in ChatGPT, Gemini and Perplexity. We then used a fresh Perplexity chat to ask it again. These were four saved manual sessions. They were not a controlled test of which product is best.

This was a small teaching check about how to inspect an answer. It was not a free customer report or a paid Full Audit. It does not use the commercial question counts. The source record is linked at the end so the example can be traced back to what was saved.

What the four sessions showed

Four saved observations, September 16, 2026
SessionWhat was visibleWhat was not established
ChatGPTActivity said 5 websites. The answer named 6 practices.The panel we checked did not show an exact list of search queries.
GeminiThe answer named 4 practices and included citations.The panel we checked did not show an exact list of search queries.
Perplexity, first run39 displayed sources and 7 search entries.The entries were not a record of real patient searches.
Perplexity, repeat10 displayed sources and 1 search entry.The product labels did not tell us the underlying model.

In the first Perplexity run, the seven search entries included our original question plus six more entries. The repeat used a fresh chat with the same visible product labels. We did not establish that every other setting was held fixed.

A displayed source count is just that: the count shown by the tool at the time. It does not mean every source was a unique practice page, correct or useful. The ChatGPT activity count and Perplexity source count are different labels. They should not be put in one chart as if they measured the same thing.

What 39 and 10 mean for a practice owner

Suppose an agency shows that an AI answer names your practice. That is worth recording. It does not show that the tool will always name you, that patients read the answer, or that the mention led to a booked visit.

A screenshot without the question leaves out part of the evidence. “Find a dentist” and “Does this named practice offer implants?” ask different things. The place, date and fresh or existing chat can also matter. Ask to see those details before treating a mention as a result you can rely on.

A better question for your agency

“Show me the exact question, date, saved answer and cited pages. Which parts did you verify against our practice’s own facts? What decision does that evidence support?”

This does not mean every answer needs a large research project. It means the strength of the claim should match the check. One saved answer can show what happened once. It cannot support a claim about a whole market.

Read the cited page before you trust the claim

In the saved research, a directory-based mention of an implant provider conflicted with the practice’s own website. We kept that conflict as something to check. We did not treat the AI answer as proof that the practice provided the service.

For your practice, use the same process. Read the cited page. Find the exact fact it supports. Compare it with a current source the practice controls, then ask the team to resolve any gap. The wrong fact may sit in an old listing rather than on the treatment page.

A useful brief says where the wrong or unclear fact appears, what it says, which source confirms the current truth, and who can request the correction. “Improve our AI” is too broad to assign to a person or check when done.

How to repeat a small check usefully

  • Choose one clear question and keep its wording. Record the place named in it.
  • Note the tool, visible mode, date and whether the chat is fresh.
  • Save the answer and any sources the interface shows. Mark missing data as missing.
  • Open the cited pages and check the claims about your practice.
  • Repeat in a fresh session. Record changes without guessing why they happened.
  • Keep observations separate from the action you propose.

Use public practice facts for this exercise. You do not need patient records or details from a private case. A source review should help check what strangers can see about the practice.

If a result is not available, record that. Do not replace it with a guessed count. If the tool changes its interface, note that too. A clean comparison depends on knowing what was actually visible, not forcing every run into the same box.

What this small sample cannot tell us

Four sessions cannot tell us how often answers change or how all dental searches work. They cannot rank these tools, tell us which one patients prefer, or prove that a page edit causes a mention. They also cannot measure how many people want implants in South Austin.

A wider study would need a fixed method, more questions, more repeated runs and clear rules for counting. The proposed larger test in the research remains a plan. We have not turned it into completed data on this page.

For now, use this example to improve the questions you ask of a report. Keep the source trail. Check the facts. Ask what was learned and what still needs testing. That is more useful than treating a changing number as a score for the practice.

The underlying evidence and replay notes are in the September 16 research record and saved demo. For the separate paid service scope, read our method.

Sources and notes

Written by Findable Practice Research for practice owners and managers. Updated 15 October 2026. Sources are linked beside the points they support. Sample wording and worked cases are labeled where they appear.

Related reading: our method, the library and company facts.