What should an AI search visibility RFP ask a vendor?
Broadcastwell measures and fixes whether AI search engines name B2B software companies.
Run the free checkAn AI search visibility RFP should ask each vendor how it measures, what it delivers by when, what it refunds, and what it will not promise, and it should accept only answers you can check. Below are 14 questions for any agency or tool, each with what a checkable answer looks like. Under each one we link our own published answer, so you can test us first.
Updated 11 November 2026.
A checkable answer names a number with its base and date, a deliverable with its clock, or a term you can read before you pay. An answer you cannot check is a claim, however confident it sounds. These questions do not rank vendors; they help you compare written answers side by side.
Measurement
1. Which engines do you measure, and do you report each one separately?
Checkable answer: a list of engines by name, one result per engine, and the capture dates. Google's two answer products behave differently, so a vendor should say whether it records Google AI Overviews and Google AI Mode apart.
Our answer: we measure ChatGPT, Claude, Perplexity, Google AI Overviews and Google AI Mode, record the two Google products separately and give identical wording to all five, per our method page.
2. Who chooses the buyer questions, and can they change mid-engagement?
Checkable answer: the question list is agreed with you before capture and frozen. If a question changes, the old comparison ends and a new baseline starts.
Our answer: the method states exactly that. Our free ten-question buyer bank shows how questions are written without brand names.
3. How many times do you ask each question on each engine?
Checkable answer: a run count per engine and question, and planned answers shown beside valid answers. One answer is an observation, not a rate.
Our answer: three scheduled runs per question and engine; an Agency Deadline Audit plans 150 answers from ten questions. Our note on measurement noise explains why one run misleads.
4. What is the base of every number in your report?
Checkable answer: every rate comes with its count, its denominator and its capture date, and an interval once the base is large enough to support one.
Our answer: below 30 observations we print counts only; at 30 or more we add a Wilson 95 percent interval. Our guide to the three measured rates defines each base.
5. What happens to answers that fail to capture?
Checkable answer: empty, error and refused answers are removed from both the count and the base, and listed separately with their reason. A failure scored as a miss makes a company look absent when the engine simply did not answer.
Our answer: our method removes them from numerator and denominator and lists each exclusion by engine and question group.
6. Can we read the answer behind each count?
Checkable answer: a receipt per answer with the engine, the date, the full text and the cited URLs, delivered with the report rather than on request.
Our answer: the fictional Kalvenor sample (Sample data, fictional company) shows every answer and receipt in the format a client receives.
7. Do you combine measures into a single score?
Checkable answer: named rate, citation rate and share of voice reported separately, each with its own base. A composite score can rise while the measure you care about falls.
Our answer: we report no composite score. Our AI Source Standard explains why visibility and attribution stay apart.
Delivery and terms
8. What exactly do we receive, and when is it due?
Checkable answer: a named deliverable, a clock with a defined start, and a recorded due time. "Within a few weeks" is not a clock.
Our answer: our operating framework quotes each clock, from the live question delivered within 24 hours of payment to the Fix Sprint test at day 30.
9. What happens if you miss the deadline or the deliverable?
Checkable answer: a refund term in writing before payment, with the amount and the time it takes to reach you.
Our answer: Answer Repair costs $490 once. If we miss the deliverable or deadline, we initiate a full $490 refund within five business days. You keep the work. You may also request a full refund within 30 days of delivery. Every refund term is quoted in the operating framework.
10. What do you promise about the AI answers themselves?
Checkable answer: no promise that an engine will name or cite you. If an outcome is tested, the test is defined before payment and measured again the same way.
Our answer: we cannot make an engine name or cite you. The Fix Sprint, $3,900 ($3,410 after the $490 credit), agrees one primary test before payment and measures it again at day 30. Measured gain on the agreed test, or the full $3,900 back. You keep the work.
11. How do you check whether the work changed anything?
Checkable answer: the same questions, engines and method version before and after, with the before and after counts side by side, each on its own base. A change is claimed only where the intervals separate.
Our answer: our completed-case format shows each field of a finished re-check on a fictional sample, and our release annotation rules record provider model releases that fall between the two captures.
People, ownership and conduct
12. Who reviews the work, by name and role?
Checkable answer: a named reviewer with a stated role and a review window.
Our answer: Priya Nair, Management Consultant, is our delivery lead and reviews each Audit within 24 hours of delivery. Our media kit lists the team.
13. Who owns the pages, the data and the exports?
Checkable answer: ownership stated in the terms, with an export of your answers, receipts and pages available to you.
Our answer: you own the domain, content and exports. Our service model lists the outputs of each workstream.
14. What will you never do, and will you publish our results?
Checkable answer: a written list of refused tactics, and a rule that your results are published only with your written approval.
Our answer: we never plant hidden text, fake reviews or instructions to models, under our Integrity Standard. Client baselines, findings, question sets and outcomes need written approval before publication.
How to use the answers
Ask every vendor the same 14 questions in writing and read the answers in one table. Where an answer gives no base, no date or no clock, ask once more for it. Our local pages standard and our Vendor Facts File show two more rules a vendor can publish for you to hold it to.
Sources
- Broadcastwell, Methodology, method v1.1, read 9 November 2026.
- Broadcastwell, Pricing, Answer Repair and Fix Sprint, read 9 November 2026.
- Broadcastwell, Integrity Standard, read 9 November 2026.
Related: the content brief template turns an agreed question into a page plan. Our vendor comparisons show these questions applied to named alternatives, each dated, and our AI search glossary defines the terms a vendor's answers should use.