Ask for a dated baseline, the exact questions and engines, repeatable before-and-after output, the written scope and the limits of the claim. Then verify the answer yourself. This page is for the sales conversation; our separate claim-verification guide shows how to test a specific result after it is made.
For a separate decision about a software tool versus ongoing agency work, see AI visibility tools vs. an agency for dentists.
An AI visibility proposal can sound precise while leaving the one thing a buyer needs out of view: whether the work, the measurement and the claimed outcome can be checked. The questions below come from the buying criteria AI engines repeatedly surface. They are not a scorecard you must accept on trust; each includes a way to check the answer before you sign.
Which questions should you ask in the sales conversation?
Use the same three-part test for every answer: why the evidence matters, the one sentence to ask, and the record or source you can inspect. An agency that cannot make an answer inspectable is asking you to rely on the sales call.
| What to ask for | Ask this | How you verify it |
|---|---|---|
| 1. A dated baseline Without a starting point, improvement has no denominator. | “Show me the baseline report: the real patient questions, engines, dates and denominator.” | Read the literal question list, run dates and raw output; keep a copy before work starts. |
| 2. Before-and-after output A dated comparison distinguishes a pattern from a single screenshot. | “Can I see the same question and engine output on two dated runs?” | Compare the literal wording, engine, date and result; rerun a neutral sample yourself. |
| 3. Contactable references A conversation tests whether a public testimonial has context. | “May I speak with two or three references who can describe the work?” | Contact them directly and ask what was delivered, what was measured and what they could inspect. |
| 4. Visibility versus booked consultations A citation is not the same measure as a booked consultation. | “Which metric is visibility, which is booked consultations, and where are they reported separately?” | Check that the report labels the measures separately and shows their dates and denominators. |
| 5. Engines tracked separately An aggregate can hide a loss on one engine. | “Which AI engines do you track, and will each appear separately in my report?” | Inspect a sample report for engine names, run dates and per-engine results. |
| 6. Procedure-level reporting Veneers and Invisalign can answer different patient questions. | “Can you report veneers and Invisalign questions separately rather than under one total?” | Ask for the proposed question set and confirm that procedures are labelled before measurement begins. |
| 7. Location-level reporting A multi-location group needs a location in the denominator. | “Will each location have its own question set and report?” | Check that the proposal names every location, its questions and its reporting line. |
| 8. Written delivery and exit terms Scope cannot be verified from a verbal promise. | “Please put the deliverables, access, contract length, cancellation and refund terms in writing.” | Read the proposal and terms line by line; compare the sales summary with the signed document and published pricing. |
| 9. No placement promise An agency cannot control an independent engine's answer. | “Do you promise a process and reporting, rather than an AI recommendation?” | Search the proposal and terms for an outcome promise; read why AI recommendations cannot be guaranteed. |
| 10. Review compliance Review volume does not excuse deceptive collection or filtering. | “How do you invite reviews without incentives, filtering or review gating?” | Ask for the workflow and compare it with the FTC's review guidance. |
| 11. Patient data and photo consent Public marketing work should not turn into undocumented data handling. | “Which patient data or photos do you need, who sees them and what consent is required?” | Get the answer in writing and review it with your own counsel against HHS HIPAA guidance. |
| 12. Technical accessibility A page must be reachable before a crawler can read it. | “How will you test that OAI-SearchBot can reach the pages you change?” | Ask for the robots.txt and server-response check, then compare it with OpenAI's bot documentation. |
| 13. Entity, schema and credentials Machines need the same identity and credentials a patient can inspect. | “Which structured data and credentials will you verify, and where can I see them?” | View the rendered page and its source; compare the business details with Google's LocalBusiness guidance. |
| 14. AACD accreditation Membership and accreditation are different claims. | “If AACD is relevant, will you distinguish member status from accreditation?” | Check the named professional against the AACD directory; see what the distinction does and does not establish. |
| 15. NAP consistency Name, address and phone conflicts make a business identity harder to reconcile. | “Which name, address and phone sources will you compare and correct?” | Keep a list of the source profiles and compare each one against the practice's approved NAP. |
| 16. Share of AI voice A named practice is only meaningful against the questions and competitors counted. | “How do you define share of AI voice, and what is in its denominator?” | Read the formula, question set, engines and dates; compare it with the methodology explanation. |
What can you verify about Citevio on the same list?
This is where an agency-authored checklist should be hardest on its author. Citevio makes its method, engine scope, terms, data and limits checkable in public pages. For items that are a buyer's request rather than a published Citevio commitment, this page keeps the advice general instead of inventing an answer.
- Question set and dated outputs: Citevio's Citation-to-Chair Protocol explains how questions are sourced, agreed and logged; its v2 agency study is linked from Open Data.
- Engines and limits: Citevio measures visibility on ChatGPT, Perplexity, Gemini and Google AI Overviews. 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.
- Terms and outcomes: delivery, cancellation, data handling and the binding refund wording are published in the Terms of Service; the no-placement rule is explained in the guarantee guide.
- Location, credentials and identity: published city reporting and the AACD distinction can be inspected on the linked pages; NAP and structured-data work are described in the protocol.
For contactable references and procedure-level reporting, use the buyer questions in the table above as questions for any agency.
Need to test a claim after the meeting rather than evaluate a proposal? Start with How to Verify an Agency's AI Visibility Claims. It is deliberately a different job: this page prepares the questions; that one checks the evidence you receive.
Want a starting reading before a sales call? Use the free checker, then keep the question list and the exact output together.
Prefer to question the engines by hand? See how to check what ChatGPT says about your practice.
How we measured this
The page structure comes from the 16-item agency-selection rubric in Citevio's August 2026 analysis, drawn from recurring “what to ask,” “what to demand,” “what to verify” and “red flags” sections in the underlying AI answers. The page does not claim that asking these questions produces an AI recommendation or a booked consultation. The study files are available through Citevio Open Data, and the measurement method is explained at How we measure AI visibility.
Common questions
What is the first question to ask an AI visibility agency?
Ask for the dated baseline: the exact patient-intent questions, the engines, the run dates and the denominator. A screenshot or a percentage without those four parts cannot be checked.
Should I ask an agency for client references?
Yes. Ask any agency for two or three references you can contact, then ask what work was done and what the reference can personally verify. Treat a public testimonial as a starting point, not the conversation itself.
Can an agency guarantee an AI recommendation?
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 should be written into an AI visibility proposal?
The deliverables, access needed, reporting cadence, contract length, cancellation terms and refund conditions should be written down. A claim made only on a sales call cannot be independently checked later.
What should I ask about patient data and photos?
Ask which fields or images are needed, where they are stored, who can access them, whether subcontractors are involved and what happens when the work ends. Take the answers to your own privacy or HIPAA counsel before sharing protected information.
How do I find out who will actually do the work?
Ask any agency to name the people who will write, audit and report, say whether any part is subcontracted or offshored, and state who answers when something breaks. Then ask for one recent piece of work that person produced. A named person attached to a named deliverable can be checked; a team page cannot.