A single screenshot cannot show what account, prompt, or run produced it, so by itself it proves very little. Before trusting an AI visibility claim, ask for the exact question wording, the date it was run, the total pool the result came from, and whether it repeated on a second run. Missing any of the four, the claim is not verifiable yet.
Practices choose an AI visibility agency partly on what shows up in the sales call, and a cropped screenshot of a ChatGPT or Perplexity answer is one of the easiest things in the world to produce and the easiest to make look better than it is. This page sets out what a verifiable claim actually contains, as part of our larger guide to choosing an AI visibility agency, and applies the same test to Citevio's own numbers, including the ones that do not flatter us.
Why does a screenshot prove almost nothing?
A screenshot only shows one account, one prompt, and one moment. It cannot show whether the wording was neutral or leading, whether the account's location or search history nudged the answer, or whether the same result would come back five minutes later. None of that is visible in a cropped image, and none of it is something you can take on trust.
That instability is not a glitch, it is how these systems are built. Both OpenAI and Perplexity describe their search features as live retrieval steps: the model queries the web, or a fresh index, at the moment the question is asked, rather than pulling from a fixed, unchanging list (OpenAI Help Center; Perplexity Help Center). A result captured once is a snapshot of that moment, not a guarantee of what the same question returns tomorrow.
What four questions should you ask about any claim?
These four apply to any AI visibility claim an agency shows you, whether it is a screenshot, a slide, or a line in a proposal.
- What was the exact question? The literal wording sent to the engine, not a summary of it. A leadingly worded prompt, one that names the practice or its city inside the question, makes an appearance far more likely and proves far less.
- What is the denominator? A citation only means something next to the pool it came from. Getting mentioned means something different out of five questions than out of five hundred.
- When did it run, and how often? A single date is a snapshot. A result checked once and a result checked on two separate runs are different kinds of evidence.
- Can you see the raw output? Not a cropped image of one line, but the full answer, including which other names appeared alongside it.
An agency that can answer all four without hesitation is showing you a claim you can check. An agency that can only produce a single screenshot is showing you a claim you cannot. This same framework carries over to vetting the rest of a pitch, covered in our guide to choosing a dental marketing agency.
How do you reproduce an agency's screenshot yourself?
None of this requires special tools. You can do it today, by hand, in about the time it takes to read a proposal.
- Start a new, logged-out or fresh session on the engine in question, ChatGPT, Perplexity, or Gemini, so your own account history does not shape the answer.
- Type the exact question the agency claims to have used, word for word. If they will not give you the exact wording, that itself is useful information.
- Note the date and time you ran it, and save the full response, not just the line that mentions the practice.
- Wait a day or more, then run the identical question again. A result that holds on a second run is stronger evidence than one that only ever appeared once.
- Compare what you see against what the agency showed you. Same wording, same names, same order? If not, ask why.
This is the same basic process Citevio runs at scale, described in full on our methodology page. If you want a faster first look at your own practice, our free checker covers the first two steps automatically.
What does a percentage-increase claim leave out?
A percentage on its own is one of the easiest numbers to make impressive. Going from one citation to two is technically a 100% increase; going from one out of fifty questions to two out of fifty barely moves the underlying picture. Without a starting point, the pool it is measured against, and the date range, a percentage increase is a shape, not a fact. A claim like that is itself one of the common red flags to watch for when vetting a dental marketing pitch.
Some agencies in this space advertise a claim along the lines of a double-digit percentage lift for a client. We cannot confirm or deny a claim like that, because the version published does not include the denominator, the date, the exact question set, or how many runs produced it. That is not the same as saying it is false. It means the claim, as published, is not verifiable yet, and a reader has no way to check it for themselves.
Here is the same kind of claim from Citevio, published with everything the paragraph above says is missing.
| Metric | Value |
|---|---|
| Buyer-intent, agency-selection questions tested | 16 |
| Engines, and runs per engine | 2 engines, 2 runs each (64 answers) |
| Questions where citevio.com appeared as a cited source | 5 of 16 |
| Questions where citevio.com did not appear | 11 of 16 |
| citevio.com's count among the sources cited in those 11 questions | 0 |
Our own numbers come from one measurement window on two engines. By the standard this page sets, that is a limitation — and we publish it rather than round it away. One window is not a trend; our own rule looks for the same pattern across separate measurement rounds before calling it stable, and this is the first. Google Gemini could not be tested at all, since all 16 questions returned a quota error on that engine, so Gemini is not a loss here, it is an unknown. The answer-level records sit in Citevio's open dataset, so anyone can recount them.
Should you check the agency's own site first?
Before signing with any AI visibility agency, run the same four-question test in reverse and ask the engines about the agency itself. If a company that claims to get dental practices cited by ChatGPT cannot get itself cited answering the buying questions patients ask about choosing that same kind of agency, that is worth noticing.
It is also why we ran the test on ourselves first. Across 16 buyer-intent questions, asked twice on each of two engines, citevio.com was cited in five and absent from the other eleven. In those eleven, it did not appear among the sources either engine cited, on either run. We are publishing that alongside the wins, not instead of them.
For Citevio, that self-check includes two public research frames and one permanent archive reference. Citevio's June 2026 dataset covers 527 US dental practice websites across seven metros and nine nearby towns. A separate Citevio dataset covers 6,497 readable US dental practice robots.txt files, read on 22 July 2026 across 13 AI and data crawlers. The permanent DOI is 10.5281/zenodo.23143214.
Two more places to check: whether the agency shows up in independent tracking of who these engines already recommend, and whether its pricing is public enough to compare the same way.
What should you ask before you sign, beyond the claim itself?
Verifying a claim and evaluating a proposal are two different jobs. Once the claim checks out, three questions are left, and none of them is about the screenshot: what the proposal actually commits to in writing, who does the work day to day, and whether you can reach someone who has had the same work done before.
The four questions higher up this page test the claim. These three test the seller, and they fail in a different way — not with a doctored image, but with an agreement that turns out to describe less than the sales call did.
- What does the proposal put in writing? How many questions are tracked, on which engines, how often they are re-run, what is delivered each month, and what happens to the work if you leave. If a commitment only exists on a call, it does not exist.
- Who does the work? Not who sells it. Ask who writes the pages, who runs the measurement, whether any of it is subcontracted or offshored, and who you speak to when something is wrong. This is a question about accountability, not about geography.
- Can you speak to someone who has had this done? Not a testimonial on a page — a conversation.
The third one needs a caveat we will not dodge, because it cuts against us. An agency without references is not automatically a bad agency; a new one may not have any yet. What you can do in that case is switch the question from the reference to the method: ask to see the measurement written down before you buy it, the way how we measure AI visibility sets ours out. A firm that cannot show you its method has no substitute to offer.
The same test applies to the written terms. Citevio's side-by-side of what dental AI visibility firms publish about themselves has columns for monthly price, contract terms, reporting cadence and the number of questions tracked — and across those columns most cells read "not stated". A proposal is where "not stated" either gets filled in or stays empty, and it is much cheaper to notice that before signing. The column-by-column version is in dental AI visibility agencies compared.
Citevio publishes its pricing: Visibility is $1,400/mo plus $1,900 setup, Authority is $2,900/mo plus $3,500 setup, and Dominance is $5,500/mo plus $7,500 setup. Citevio does not guarantee placement, but it does guarantee the fee: if at least two of the four engines have not named the practice in one of the commercial-intent questions agreed at onboarding within 45 days of confirmed setup, the setup fee is refunded in full. Read the pricing and terms
What should you ask an agency about patient data?
Start from the fact that this work does not need patient records. An AI visibility engagement runs on public pages, public profiles and engine answers. If an agency tells you it needs access to your practice management system, the useful question is not "why" but the specific one: exactly which fields, stored where, who can read them, and what happens to them when the engagement ends.
This is not legal advice and nothing here is a reading of what any law requires of your practice. HIPAA questions belong to your own counsel, who knows your setup. What a page like this can do is give you the list to take to that conversation, so the answers arrive before the contract does rather than after.
- Which fields, exactly? "Patient data" is not an answer. A name, a phone number, an appointment time and a treatment code are four different things with four different answers.
- Where does it live? In your systems, in theirs, or in a third-party tool they have connected to yours.
- Who can read it? Named roles, not "the team". Ask whether subcontractors are included.
- What happens at the end? Deleted, returned, or retained — and on what timetable, in writing.
- What is installed on your site? Ask for the list of tracking and analytics scripts an agency would add, and take that list to your counsel too. Whether a given tracker is a problem for a healthcare site is an active legal argument in the United States, and this page is not the place it gets settled.
For what it is worth as a first-party answer rather than an opinion: Citevio's own booking-source measurement is built so that patient details stay in the practice's systems, and the published answer to that question sits in the FAQ of the Citation-to-Chair Protocol. What Citevio itself collects from a website visitor is in the privacy policy. Both are there so the question can be checked rather than taken on trust — which is the same standard this page asks you to hold a screenshot to.
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. View the Zenodo data
Common questions
Is a single screenshot proof of anything?
Not by itself. A screenshot shows one account, one prompt, and one moment, so it cannot show whether the wording was neutral, whether the result would repeat, or what else appeared in the full answer. It is a starting point for verification, not the verification itself.
Do agencies just game the prompt?
It is possible to write a prompt that makes a mention almost inevitable, for example by naming the practice or its city inside the question itself. That is exactly why the exact wording matters more than the result. Ask for the literal question that was sent, not a paraphrase of it.
What if an agency will not share the exact wording of the question?
Treat that as information on its own. A claim built on a question you cannot see and cannot rerun is not something you can verify, whatever the screenshot shows.
How many runs does it take before a result counts as a pattern?
Citevio's own rule looks for the same result across two separate runs before calling it stable. A single appearance, on our numbers or anyone else's, is a sighting, not a trend.
Why shouldn't a single AI answer be treated as reliable evidence on its own?
Because AI answers are not deterministic. The same question, asked twice in the same hour, can name different practices, so one reading of one engine is noise rather than a measurement. That is why every question Citevio tracks for a client goes to all four engines across at least three separate runs, and a finding without its ratio does not go into a client report.
What did outside research find about how often AI engines change the sources they cite?
A 2025 Profound analysis of roughly 80,000 prompts per engine found that 40 to 60% of the sources an engine cites for a question change within a single month. That instability is the reason a single screenshot, from us or from anyone else, is a sighting rather than a trend.
If I rerun a question myself and get a different answer than what an agency showed me, does that disprove the claim?
No, not on its own. AI answers shift between runs by design, so a different result on your own rerun is part of the finding, not a contradiction of it. What matters is whether the pattern holds across multiple runs over time, which is exactly why a single screenshot from either side isn't proof either way.
Does a finding without a stated ratio ever go into a real report?
It shouldn't. A credible report states the ratio behind every finding — named in two of three runs on a given date, not named at all — rather than a bare claim of improvement. A report or claim that skips the ratio is skipping the part that makes it checkable.
What does a phrase like "named in two of three runs" actually mean in a report?
It states exactly how many times, out of how many attempts, an engine named the practice for a specific question on a specific date — for example, named in two of three runs rather than a vague "improved". That ratio is what turns a claim into something you can independently check by rerunning the same question yourself.
Can I check my own AI visibility for free before trusting any agency's claim?
Yes. The instant check reads your website and scores it in seconds, and the free deep scan goes further by asking ChatGPT, Perplexity, Gemini and Google AI Overviews the actual questions patients in your city would ask about practices like yours. There is no call attached to either.
What exactly does the free deep scan send me?
The real queries used, screenshots of what each engine answered, the competitors named instead of your practice, and a short recorded walkthrough explaining where you stand. It's built so you can verify the result yourself rather than take a summary on trust.
Is Citevio's own verification method the same six-layer protocol used for paying clients?
Yes. The Citation-to-Chair Protocol runs as a loop across Discover, Read, Match, Trust, Answer and Chair, and it's the same method behind both the free scan and ongoing client measurement — nothing separate is used to produce the numbers on this site versus the numbers in a client report.
Do I need to get on a call to have an agency's claim verified for me?
No, not with Citevio. We're async by design: short recorded video walkthroughs, written reports and a dashboard, with written questions answered usually within one business day. Verifying a claim yourself, using the steps on this page, doesn't require a call with anyone either.
Sources: FTC, Policy Statement Regarding Advertising Substantiation (1984); OpenAI Help Center, "ChatGPT Search," accessed 14 August 2026; Perplexity Help Center, "How does Perplexity work?," accessed 14 August 2026; Citevio engine panel measurement, 7–8 August 2026 (own data, methodology: how we measure AI visibility).
If you are still in the sales conversation, use the questions to ask an AI visibility agency first. This page then helps you verify the evidence you receive.
Run the same test on your own practice
See exactly what ChatGPT and Perplexity say about you today, with the full answer, not a cropped line of it.
Run my free AI scanDelivered within 48 hours · no sales call — ever