In Citevio's run-consistency measurement on 12 August 2026, a two-run check across 16 buyer-intent agency-selection questions, not patient-facing dental queries, produced 64 valid answers from ChatGPT and Perplexity. ChatGPT gave the same headline answer on the flagged query in both runs. Perplexity's headline recommendation changed between runs. Two runs show instability exists; they do not tell you how often it happens.
A changed answer is not automatically evidence that something broke. This measurement tests a narrow question: whether repeated runs of the same buyer-intent agency-selection questions produced the same headline answer. It does not establish a change rate or explain an individual answer.
What did we find when we asked the same question twice?
In Citevio's run-consistency measurement on 12 August 2026, we took the same 16 buyer-intent agency-selection questions and ran each one twice on ChatGPT and twice on Perplexity. That produced 64 valid answers in total. These were agency-selection queries, not patient-facing dental queries, so this is a narrow read on repeated engine answers rather than a study of dental search.
| Engine | Runs | Result |
|---|---|---|
| ChatGPT | 2 | Stable: same headline answer on the flagged query in both runs |
| Perplexity | 2 | Unstable: headline recommendation changed from run 1 to run 2 |
Why does the same question give a different answer?
We did not measure the cause of the difference. Search-backed AI answers can draw on retrieval and ranking systems, but this measurement was not designed to isolate which system behavior changed an answer. It records the observed pattern only.
Why might I appear once and then disappear?
If a practice is named once and not named on a later check, the two-run result shows that a change is not automatically evidence that the practice did something wrong. Citevio has not measured which explanation applies to any individual case; that requires a targeted check of the specific question.
How do you tell a real change from noise?
In Citevio's run-consistency measurement on 12 August 2026, a single run was one data point. We used two runs per engine as the minimum check for whether a headline answer repeated or changed. Even two runs are not a change-rate measurement, so they are a basis for further checking, not a forecast.
Citevio's run-consistency measurement on 12 August 2026 used two runs per engine. They can show that an answer is unstable, but they cannot tell you how often it changes, and we do not publish a change rate we have not measured.
That scope is also separate from Citevio's site-anatomy baseline: 527 US dental practice websites across seven metros plus nine nearby towns, read field by field between 6 and 26 June 2026.
Do Google AI Overviews change the same way?
We measured run-to-run consistency on ChatGPT and Perplexity only. Our Gemini run returned a quota error on all 16 questions, and we have not measured AI Overviews, so for those two the honest answer is that we do not know. Google documents AI Overviews as a Search feature, but that does not tell us whether its output is more or less stable.
How long does this take, and why we will not give you a date
We will not give you one, and you should be wary of anyone who does. No timeline for appearing in an AI answer has been measured here, so any date offered would be a guess with a confident face on it. What can be offered instead is a check schedule: after something is published, the same question is asked again on the same engine under the same conditions, at fixed intervals, and the results are kept with their dates.
The reason to trade the first question for the second is not modesty. It is that we have run exactly this kind of check on ourselves and published the result when it went against us.
On 16 August 2026 Citevio ran a three-day follow-up on one of its own published posts: one query, one engine, a single run on Perplexity. The query returned twenty sources. None of them was ours. The baseline for the same query five days earlier was also zero, so nothing had moved. That is a single run and is described as one everywhere it appears, because a single run cannot establish either success or failure — which is precisely why it does not license the conclusion that publishing there does not work.
Source: Citevio follow-up measurement, 16 August 2026, one query on Perplexity, single run, twenty sources returned. Baseline for the same query, 11 August 2026: zero.
Two other measurements explain why a date would be meaningless even if we had one. Across three fixed queries run twice on each of two engines through their APIs on the evening of 15 August 2026, with 12 of 12 calls completed, the displayed order moved between runs — on one engine the same query came back first in one run and third in the next, which is one evening's photograph rather than a trend. And in the run-consistency work, one engine repeated its headline answer across two runs while the other changed it. When the answer moves inside a single evening, a promise about a particular week is not a forecast, it is a sales device.
So what does a schedule look like in practice? A recorded baseline before anything changes, the same questions re-asked at fixed intervals afterwards, every reading stamped with its date, engine and run number, and a rule agreed in advance for what counts as a move rather than noise. That rule is the part most reporting skips, and it is set out separately in the section above on telling a real change from noise.
One question deserves a direct non-answer. If a practice changes nothing and simply waits, will it appear eventually? We have not measured that. No cohort of practices has been tracked over time with no intervention, so there is no honest yes and no honest no — only an absence of evidence that nobody should fill in with a hunch. What Citevio publishes about its own visibility over time, wins and losses together, is in the Citevio AI visibility case study; the sampling rules behind every reading are on how we measure AI visibility; and what is actually being paid for month to month, given that no date is on offer, is set out on what GEO for a dental clinic costs.
Which questions are commonly asked about changing AI answers?
How do I keep my site fresh so AI engines have something current to draw on?
There is no run-to-run consistency data tying update frequency to answer stability. Keep important facts current and available in text, but do not treat a content update as a guaranteed way to change an AI answer.
Does a different answer mean I did something wrong?
Not necessarily. In Citevio's run-consistency measurement on 12 August 2026, ChatGPT gave the same headline answer on the flagged query in both runs, while Perplexity's headline recommendation changed between run 1 and run 2. The measurement shows a change can occur; it does not identify the cause in an individual case.
Can you tell me how often my listing will change going forward?
No. Citevio's run-consistency measurement on 12 August 2026 used two runs per engine; those runs can show that an answer is unstable, but they cannot tell you how often it changes. We have not measured a change rate on any engine, and we do not publish one we have not measured.
What exactly did you test in the run-consistency measurement?
Whether the same question, asked twice on the same engine, produces the same headline answer. On 12 August 2026 we ran 16 buyer-intent agency-selection questions twice each on ChatGPT and Perplexity, producing 64 valid answers. These were agency-selection queries, not patient-facing dental questions.
Did ChatGPT or Perplexity change its answer between the two runs?
Perplexity did; ChatGPT did not. On the flagged query, ChatGPT gave the same headline answer in both runs, while Perplexity's headline recommendation changed from run 1 to run 2.
Do you know why Perplexity's answer changed and ChatGPT's did not?
No. This measurement records whether the headline answer repeated or changed; it was not designed to isolate which underlying retrieval or ranking behavior caused the difference. That is an open question, not a finding we have.
Was Gemini or Google AI Overviews part of this test?
Gemini was attempted and failed: our run returned a quota error on all 16 questions. Google AI Overviews were not measured at all. For both, the honest answer is that we do not know how stable their answers are.
Does a two-run test actually prove anything?
It proves that instability can exist, and no more than that. Two runs per engine is the minimum check for whether a headline answer repeats or changes; it is a basis for further checking, not a change-rate measurement, and we do not extrapolate a frequency from it.
If my practice was named once and then was not, does that mean something is wrong with my listing?
Not necessarily, and we cannot tell you which explanation applies without checking the specific question. The run-consistency result shows that a change can happen on its own, without anything about a practice changing, so a single disappearance is not automatically evidence of a problem.
What is an example of AI answers moving within the same day?
On the evening of 15 August 2026 we ran three fixed queries twice each on two engines through their APIs, with 12 of 12 calls completed, and the order of results moved between runs. On one engine, the same query returned a source first in one run and third in the next.
Have you tested whether your own content gets cited by AI over time?
Yes, and we published an unfavorable result. On 16 August 2026 we ran a single-query, single-run check on Perplexity, three days after publishing a post; it returned twenty sources and none were ours. The baseline for the same query five days earlier was also zero, so nothing had actually moved.
Why do you keep saying "single run" about your own results instead of just giving a verdict?
Because one run cannot establish success or failure on its own, which is the whole point of this page. A single Perplexity check returning zero mentions of us is a data point, not proof that publishing there does not work; it needs to be repeated on a schedule before it means anything.
If I wait long enough without changing anything, will my practice eventually show up?
We do not know, and we are not going to guess. No cohort of practices has been tracked over time with zero intervention, so there is no honest yes and no honest no on offer here.
Can you tell me exactly when my practice will start appearing in ChatGPT?
No, and you should be cautious of anyone who does. No timeline for a practice appearing in an AI answer has been measured on this page or anywhere else on our site; any date offered would be a guess dressed up as a forecast.
What can you offer instead of a timeline?
A check schedule. After something is published, the same question gets asked again on the same engine, under the same conditions, at fixed intervals, with every result kept and dated, rather than a one-time reading treated as a verdict.
How many questions and runs went into the 12 August measurement, exactly?
Sixteen buyer-intent agency-selection questions, run twice on each of two engines, ChatGPT and Perplexity, for 64 valid answers total. It is a narrow, dated measurement of run-to-run consistency, not a study of patient-facing dental search.
Do you only publish measurements that make Citevio look good?
No. The 16 August follow-up above returned zero mentions of Citevio on Perplexity, and we published it anyway. What Citevio publishes about its own visibility over time, wins and losses together, is also in the Citevio AI visibility case study.
Can I check your run-consistency numbers myself?
Yes. Citevio's dental market research, including this measurement's methodology, is published under a CC-BY-4.0 license with a DOI on Zenodo, so the numbers can be checked and reused. Client data is never part of what is published.
What counts as a real change versus normal noise in these measurements?
We used two runs per engine as the floor for telling a repeated answer from a changed one, with the rule for what counts as a move agreed in advance rather than decided after seeing the results. Even so, two runs are a basis for checking further, not a forecast of what happens next.
Does Google document how stable its AI Overviews are supposed to be?
Not in a way that answers this question. Google documents AI Overviews as a Search feature, but that documentation does not tell us whether its output is more or less stable, run to run, than ChatGPT's or Perplexity's, and we have not measured it ourselves.
Citevio's dental market research is published openly under CC-BY-4.0 with a DOI, so the numbers can be checked and reused; client data is never published.
Sources: OpenAI, ChatGPT search; Perplexity, What is Perplexity?; Google Search Central, AI features and your website; Citevio open dataset, Zenodo DOI 10.5281/zenodo.23143214; Citevio run-consistency measurement, 12 August 2026 (own data, methodology: how we measure AI visibility).
Want to see what AI engines say about your practice right now?
Get a free scan of how ChatGPT and Perplexity currently answer questions about you.
Run my free AI scanDelivered within 48 hours · no sales call — ever