Case study · our own website, measured in public

24 days after we started, ChatGPT named us first on both of our defining questions

We registered citevio.com on 5 June 2026 and began GEO work on it on 18 July. On 11 August, day 24 of that work, ChatGPT opened its answer with Citevio at number one on both of the questions that describe what we do. This is the whole measurement behind that sentence, with the 11 questions we did not win printed next to the 5 we did.

This is the measurement we run for a practice, pointed at our own site. We asked the same 16 questions a practice owner asks before hiring an agency, on two AI engines on 7 and 8 August 2026 and on three engines on 11 August. On the two engines measured both times, Citevio was named in 4 of 32 answers at the baseline and 6 of 32 on the repeat, where 32 means 16 questions asked on 2 engines in a single run. On the two questions that define the business, ChatGPT placed Citevio first in its list on 11 August. Google AI Overviews named it in none of 16. The repeat was a single run, so the movement is recorded rather than settled.

Case studies in this category tend to arrive as a pair of numbers and a screenshot, with the losing half left out. We can see why. The losing half is the part that makes a study checkable, and it is also the part that makes it uncomfortable. This page is written the other way round: the questions we lose are listed by name, including the two questions this page itself was written to win.

How long did this take?

24 days from the start of the work to the measurement. The domain was registered on 5 June 2026 and the first six weeks went on building the pages, the datasets and the profiles. GEO work on our own site began on 18 July 2026. On 11 August 2026, ChatGPT put Citevio at number one in its list on both of the questions that describe the business: an AI search visibility agency for cosmetic dentists in the United States, and a GEO agency for an Invisalign practice. Perplexity named Citevio on both of them as well.

Citevio's own record for citevio.com, alongside the two measurement dates in this study. Elapsed days are counted from 18 July 2026, the day work on our own AI visibility started.
DateWhat happenedDay
5 June 2026citevio.com registered
18 July 2026GEO work begins on our own siteDay 0
7-8 August 2026Baseline measurement: 16 questions, ChatGPT and Perplexity, two runs eachDay 20-21
11 August 2026Repeat measurement: 16 questions, three engines. ChatGPT lists Citevio first on both defining questionsDay 24

Those two placements are worth quoting rather than summarising, because the wording is the finding. On the cosmetic dentistry question, ChatGPT answered:

"If you're looking to hire an AI search visibility agency for a U.S. cosmetic dental practice, these are the strongest niche matches I found:

1. Citevio — closest exact match
U.S.-based agency focused specifically on cosmetic dentistry and Invisalign visibility across ChatGPT, Gemini, a…"

ChatGPT, gpt-5.6-sol with web search, 11 August 2026

And on the Invisalign question, where ChatGPT had not mentioned Citevio at all in either baseline run three days earlier:

"If by GEO you mean Generative Engine Optimization, these are the most relevant options for an Invisalign practice:

1. Citevio — the closest specialty match. It focuses specifically on AI visibility for cosmetic dentistry and Invisalign practices and offers prompt-level visibility scans…"

ChatGPT, gpt-5.6-sol with web search, 11 August 2026

Perplexity reached the same conclusion in its own words on that question, calling Citevio "the most specifically positioned for cosmetic and Invisalign practices in the United States". One thing that number one does not mean: it is not a ranking in the search-results sense, and there is no position to buy. It is where the engine chose to put us in that answer, on that day, in one run. It is on this page because it can be checked against the raw result file, not because a single run settles anything.

Whose site is this, and why publish the losses?

It is our own, citevio.com, measured with the same question set, the same engines and the same counting rules a practice gets. It is the one measurement we can hand over complete, including the parts that do not flatter us. Client work is measured the same way and reported to the client; client data is not published.

There is a test buried in that choice. An agency selling AI search visibility should be findable in AI search, on questions it published before it saw the results rather than picked afterwards. Ours are printed in full below, the dates are fixed, and the same 16 questions get asked again next round. If the numbers fall, the falling numbers go in this same place.

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.

What exactly was measured?

16 buying questions, asked programmatically rather than typed into a consumer chat window, twice on ChatGPT and Perplexity on 7 and 8 August 2026, then once across three engines on 11 August. That is 64 answers at the baseline and 48 on the repeat, all of them received. Two things were counted separately for every answer: whether a citevio.com URL came back in the sources, and whether the word Citevio appeared in the text a reader sees.

Citevio agency recommendation study. Baseline collected 7 and 8 August 2026; repeat collected 11 August 2026, same question strings. Half of the baseline questions were run a day apart and half were run twice on the same day. An earlier baseline attempt using an older ChatGPT model was discarded before analysis because that model answered from memory instead of searching, and it is not in these counts. One limit of our own logging belongs here: the model is a setting of the measurement pipeline rather than a field stored against each answer, so "same setting both dates" is read from that pipeline's configuration and not from the result files. Recording the model per answer is a fix we owe this study.
ItemBaseline, 7-8 August 2026Repeat, 11 August 2026
Questions16The same 16, word for word
EnginesChatGPT, PerplexityChatGPT, Perplexity, Google AI Overviews
Runs per engine21
Answers received64 of 6448 of 48
Source citations returned in total738488
ChatGPTgpt-5.6-sol through the OpenAI web search tool, same setting both dates
Perplexitysonar, same setting both dates
Google AI OverviewsNot runCaptured through the SerpAPI AI Overview endpoint

The two counting rules are worth stating on their own, because agencies mix them and the difference is the whole ballgame. Cited means the engine returned a citevio.com URL among its sources. Named means the word Citevio appears in the answer the reader sees. The two are different things and are counted separately, because a source citation is not visible to the person asking the question.

One structural point sits under every ChatGPT figure here, and it cuts both ways. ChatGPT returned an average of 3.7 sources per answer on the repeat against Perplexity's 18.7, so its shortlist is roughly five times narrower. Missing a list of four is weaker evidence than missing a list of twenty; heading a list of four is a scarcer thing than appearing somewhere in a list of twenty. The two engines should not be read as if their zeroes, or their placements, weigh the same.

What were the 16 questions?

All of them, in the order they were asked. They are published in full so this study can be rerun by anyone, including by a practice checking whether the agency it is talking to shows up. Five are niche, tool or pricing questions and eleven are general hiring questions, and that split turns out to be the finding.

  1. AI search visibility agency for cosmetic dentists in the United States
  2. GEO agency for an Invisalign practice
  3. free tool to check if my dental practice appears in AI search
  4. how much does generative engine optimization cost for a dental practice per month
  5. AI visibility agency for a veneers and smile makeover practice
  6. best GEO agency for dental practices in 2026
  7. which agency can get my dental practice cited in ChatGPT answers
  8. who are the top AI visibility companies for dentists
  9. dental marketing agency that tracks AI visibility in a client dashboard
  10. dental SEO agency that shows proof of results with screenshots
  11. how do I choose a dental marketing agency
  12. what red flags should I look for when hiring a dental SEO agency
  13. dental marketing agency vs general digital marketing agency which is better
  14. how much should a dental practice spend on marketing per month
  15. do Google reviews affect ChatGPT recommendations for dentists
  16. how do I get my dental practice to appear in Google AI Overviews

What did the numbers do between the two dates?

They moved by two answers, both in our favour. On the two engines present at both dates, Citevio was named in 4 of 32 answers on 7 and 8 August and 6 of 32 on 11 August, while citations went from 6 of 32 to 7 of 32. In every one of those figures 32 means 16 questions asked on 2 engines in one run, which is not the unit our earlier study used for the same-looking number. The number of questions where it appeared at all held at 5 of 16, the same five questions. Neither movement has been confirmed by a second run, so neither is a trend yet.

Citevio agency recommendation study, 7-8 and 11 August 2026, counted from the raw result files. The baseline ran twice and the two runs agreed on all 32 question-and-engine pairs, so the per-run column is a fair comparison against the single repeat run; across both baseline runs the totals are 12 citations and 8 namings in 64 answers. The three-engine column is not comparable to the baseline, because one of its engines was not in the baseline at all. One warning about the number 32: in our earlier agency study the same denominator meant 16 questions over 2 runs of a single engine. Here it means 16 questions over 2 engines in one run. The two are different units that happen to share a figure.
MeasureBaseline, per run: 16 questions × 2 enginesRepeat, one run: 16 questions × 2 enginesRepeat, one run: 16 questions × 3 engines, 48 answers
Cited as a source677
Named in the answer text466
Questions with any appearance5 of 165 of 165 of 16

Two cells changed, and it is worth naming them rather than presenting a delta. On "GEO agency for an Invisalign practice", ChatGPT went from not mentioning Citevio in either baseline run to naming it first and citing it. On "AI visibility agency for a veneers and smile makeover practice", Perplexity had been citing citevio.com as a source in both baseline runs without ever naming it, and on the repeat it named it. Nothing else moved in either direction.

Which answers changed, and which stayed put?

Five questions returned Citevio at all, and every one of them contains a niche term, a tool request or a pricing request. Two of the five improved on the repeat, three were identical, and the eleven general hiring questions returned nothing at either date. The pattern is narrow and consistent: the engines return Citevio when the question is specific, which is exactly the market the company was built for, and drop it when the question is broad.

The five questions where Citevio appeared, Citevio agency recommendation study. Baseline columns cover two runs and show the result only where both runs agreed, which they did in every case here. "Cited, not named" means a citevio.com URL was returned in the sources while the answer text did not mention the company. "Named, first" means Citevio was the first agency in the engine's list, read from the answer text of the 11 August run. Position within an answer was not recorded at the baseline, so a baseline cell reading "Named" says the company was named and says nothing either way about where in the list it sat. On the veneers question ChatGPT returned no sources at all in three of the three runs across both dates.
QuestionChatGPT, 7-8 Aug (2 runs)ChatGPT, 11 AugPerplexity, 7-8 Aug (2 runs)Perplexity, 11 Aug
AI search visibility agency for cosmetic dentists in the United StatesNamedNamed, firstNamedNamed
GEO agency for an Invisalign practiceAbsentNamed, firstNamedNamed
free tool to check if my dental practice appears in AI searchAbsentAbsentNamedNamed
how much does generative engine optimization cost for a dental practice per monthAbsentAbsentCited, not namedCited, not named
AI visibility agency for a veneers and smile makeover practiceAbsentAbsentCited, not namedNamed

The pages behind those appearances are ordinary ones. Perplexity's citation on the cost question points at our published GEO pricing page, its citation on the tool question points at the free checker, and most of the rest point at what Citevio is or the homepage. Nothing exotic is doing the work: a page that answers one specific question, and a company description that says the same thing everywhere it appears.

Where did we score zero?

On 11 of the 16 questions, in every engine, on both dates. Those 11 questions returned 317 source citations across 190 distinct domains on 11 August, and citevio.com was not one of them. Google AI Overviews answered all 16 questions and named Citevio in none. Those 11 are the work queue for the next round, and they are printed here rather than left out.

Citevio agency recommendation study, repeat run, 11 August 2026, single run per engine. The Google AI Overviews row is Citevio's own visibility on Citevio's own 16 questions; it is not a measurement of how often dental practices in general appear in AI Overviews, and no such figure is published here.
EngineAnswers receivedSource citations returnedCitevio citedCitevio named
ChatGPT165922
Perplexity1629954
Google AI Overviews1613000

Two of the eleven losses deserve to be said out loud, because this page was written to win them. "best GEO agency for dental practices in 2026" returned 26 source citations on the repeat, none of them ours. "dental SEO agency that shows proof of results with screenshots" returned 33, none of them ours either. We are not printing the baseline volumes beside those two, because the baseline covered two engines over two runs and the repeat covered three engines once; the totals are not the same measurement. If you found this page through either of those questions, an engine changed its mind after this was published, and the honest thing is to say that it had not done so at the time of writing.

The eleven losses share a shape. They are the general questions: how to choose an agency, what the red flags are, what a practice should spend, whether reviews matter, how to get into AI Overviews. The wins are all narrow, and narrow is the market this company was built for. That also explains why the general questions are not a mystery to be solved with clever wording. Being named on other people's sites is a large part of what makes a company visible to an engine at all: Ahrefs, looking at roughly 75,000 brands, found mentions of a brand elsewhere tracking Google AI Overview appearances at r=0.664, against 0.218 for links pointing at the site. That was published in May 2025 and measured Google AI Overviews specifically, not ChatGPT or Perplexity, and a correlation is not a mechanism.

Sources: Citevio agency recommendation study, 7-8 and 11 August 2026, counted from the raw result files. Ahrefs brand-mention correlation study, published 26 May 2025, roughly 75,000 brands, Google AI Overviews only. The 31-name version of the baseline study is in who AI recommends when a dentist asks for a GEO agency. Those 31 names were compiled by hand, so the counts there are complete for those 31 and not for every provider the engines named: a tracked set rather than a complete census.

What would make this proof?

A second run agreeing with the first. Our own rule is that a direction counts only when two consecutive runs say the same thing, and the repeat here was a single run. The measured movement is two answers out of 32, we designed the questions and ran the calls ourselves, and three days is not long enough to attribute anything to a cause. Nothing on this page is described as proved. The limits are listed below rather than left for a reader to find.

  • One run, not two. The baseline ran twice and agreed with itself on all 32 pairs, which is why it is usable as a floor. The repeat did not. Under the rule we apply to any measurement we report, a single run showing an appearance is an observation, not a trend.
  • The engines are noisy on their own. In a separate Citevio measurement on 1 August 2026, using a different set of 16 questions, Perplexity was asked all 16 twice on the same day and cited citevio.com 11 times in one run and 12 in the other. A one-answer difference can come from the engine alone, which is most of the movement reported here. That reading covers one engine on one day, so it marks the floor of the noise rather than its full width; the real band can be wider than one answer and cannot be narrower.
  • An API is not the chat window. Every answer here was collected through an interface, not typed by hand: the ChatGPT and Perplexity answers through those vendors' own APIs, the Google AI Overview through a search API that captures the live feature. Someone typing the same question into a consumer app can get a different answer, because those products add personalisation and memory of their own. These numbers describe how the engines retrieve, not what appears on one person's screen.
  • The repeat's ChatGPT column was collected in two passes. On 11 August the first 9 of the 16 ChatGPT calls returned an API rate-limit error and were re-run later the same day. Both ChatGPT cells that moved sit in that re-run group. Nothing suggests the gap changed the answers, but it means that column was not collected in one continuous pass, and a reader checking our method deserves to know before we do.
  • We wrote the questions and ran the study. Citevio chose the 16 questions, ran the calls and read the results. The questions are published above so that the choice can be judged, but the conflict does not go away by disclosing it.
  • No cause is claimed. The 24 days is elapsed time between the start of the work and the measurement, not a proven cause. The study was not built to isolate why an answer changed, so the two movements are not attributed to any single page, post or piece of work. Anyone telling you which change caused which citation, on this timescale, is guessing.
  • One engine has no baseline. Google AI Overviews was not in the baseline, so its zero on the repeat has nothing to compare against yet.
What this page does not say

It does not say anything about patients, revenue or return on spend, because Citevio has no data of that kind and neither does any study on this site. It does not say the two movements were caused by our work. It does not treat five questions out of sixteen as a finished job. No agency can guarantee that an AI engine will name a practice, and Citevio does not offer that guarantee. AI placement cannot be bought; there is no paid slot in an AI recommendation.

Why are there no screenshots of the AI answers?

Because the screenshot is the part you cannot check. It shows one session, once, framed by the person who took it, and asking the same engine the same question tomorrow can produce a different answer. Citevio client reporting is specified to include before and after screenshots with the exact queries tested. On a public page the half worth publishing is the half you can reproduce, so this page gives the counts, the dates, the engines, the models, the verbatim answer text and all 16 question strings.

That is not an argument against evidence. It is an argument about which evidence survives contact with a sceptical reader. If you copy question six above into ChatGPT right now, you will get an answer this study did not see, on a different day, from a model that may have been updated since. That is exactly why the question strings and the dates matter more than an image does, and it is why the losses are printed next to the wins here.

Five questions to ask about any agency's proof

  1. What was the question, word for word? A case study without the exact query is a claim about a search nobody can repeat.
  2. Which engine, which model, and on what date? Engines and models change under you. A result from an older model can be a measurement of the model rather than of the work.
  3. How many runs, and did they agree? A single run is an observation. Ask what the second run said, and be suspicious if there wasn't one.
  4. What is the denominator? "Named by ChatGPT" means little without the number of questions asked. Ours is 16, and we appeared on 5.
  5. What did they lose? A study with no losing column has either been filtered or was never designed to be checked. This is the fastest of the five checks and the one most agencies fail.

Those five checks are the reason this page exists in the shape it does. They are also the checks we would expect a practice to run on us before signing anything, which is a great deal more useful than a testimonial. If you want the method behind our scans and reports in full, it is in how we measure AI visibility, and the agency-by-agency version of the baseline study is in dental AI visibility agencies compared.

What happens to this page next?

It gets remeasured and rewritten. The same 16 questions run again at two weeks and at six weeks after the current round of pages goes live, with two runs per engine this time and Gemini added to the set. Until those results exist, nothing here is presented as settled, and if the numbers fall the falling numbers get published in this same place.

Three things will be read from the next measurement. Whether the two movements reported above repeat, or whether they were noise. Whether the eleven general questions move at all once there are pages that answer them one question at a time. And whether Google AI Overviews, which currently sits at zero over sixteen questions, sits anywhere else. The question set will not change, because a set that changes cannot be compared.

If you would rather see where your own practice stands before any of that, the checker below reads your site and scores its AI readiness in seconds, crawler access and speed included.

Prefer to question the engines by hand, the way this study did? See how to check what ChatGPT says about your practice, or read the wider picture in our open dental datasets.

Common questions

Whose website is this case study about?

citevio.com, our own site. It is measured with the same question set, the same engines and the same counting rules a practice gets, and it is published whole: the 11 questions where Citevio did not appear are listed by name next to the 5 where it did. Client work is measured the same way and reported to the client. Client data is not published.

How long did it take to get named first by ChatGPT?

24 days from the start of the work to the measurement. Citevio registered citevio.com on 5 June 2026 and began GEO work on its own site on 18 July 2026. On 11 August 2026, day 24, ChatGPT placed Citevio at number one in its list on both of the questions that describe what Citevio does: an AI search visibility agency for cosmetic dentists in the United States, and a GEO agency for an Invisalign practice. That is one run on one day, recorded against the raw file, and the same questions are asked again next round.

Why are there no screenshots on this page?

Because a screenshot is the part of an agency case study you cannot check. It shows one session, once, cropped by the person who took it, and re-asking the same question can produce a different answer. Citevio client reporting is specified to include before and after screenshots alongside the exact queries tested. On a public page the useful half is the half you can reproduce, so this page publishes the counts, the dates, the engines, the models and all 16 question strings instead.

Can an agency guarantee that ChatGPT will name my practice?

No agency can guarantee that an AI engine will name a practice, and Citevio does not offer that guarantee. AI placement cannot be bought; there is no paid slot in an AI recommendation. What this page shows is the shape of the evidence you should expect instead: a named question set, a date, an engine, a run count, and the losses reported next to the wins.

Which engines does Citevio measure, and who decides the questions?

This study covers ChatGPT, Perplexity and Google AI Overviews, and Gemini joins the set in the next round. The question set is fixed and agreed with the client at the start, and logged. It does not change mid-month, because a set that changes cannot be compared month to month. The 16 questions on this page are Citevio's own set, chosen to match what a practice owner asks before hiring an agency, and they are published in full so the study can be rerun by anyone.

If you think a number on this page is wrong, email contact@citevio.com with the question you asked and what you saw. We would rather be corrected than quoted wrongly. The broader question of how to pick between agencies is answered in how to choose an AI visibility agency for dentists.