Method
Your buyers already told you what they weigh.The record is in your sales calls.
Voice of Customer is our analysis of your recorded sales calls, read against the Foundation we built for you. We take one call at a time, keep every snippet with its speaker role, call, and date, and count which rows of the map your buyers confirmed, which none of them ever echoed, and which themes the map never had. Your 150 buyer questions are rewritten from what they said. It is part of the $997 AI Visibility Crawl, and it runs when you share calls.
Why it matters
Every tool in this category scores you against questions it guessed.
A visibility score is only as good as the questions behind it. Put a set of questions to four engines and you get a number, and that number describes the questions rather than your market. So the discipline rests on a step almost nobody examines: where did the questions come from?
They came from the public web. The category's tools read what has been published about an industry, infer how buyers in it probably talk, and generate prompts from that inference. Some of it is good. None of it is a record of anything a buyer said.
Your Foundation starts the same way. We build it from public evidence, which makes it a well-researched guess about your buyers. Voice of Customer is the step that turns the guess into a record. The stakes sit in the gap: 95% of winning vendors were already on the buyer's Day One shortlist, a set of four names, before sales was involved (6sense, B2B Buyer Experience Report 2025). The questions decide which shortlists you can enter.
We can put a number on how much the question shape matters. Across 8,999 executed AI answers for 22 clients, Resonate Labs measured how often an unbranded buyer question produced an actual recommendation: shortlist-format questions did it 33% of the time, comparison questions 8%, and problem-framing explainers under 5%. Recommendation share held steady at 54% to 61% across ChatGPT, Claude, Gemini, and Perplexity, which is the part worth sitting with. The question you are measured on decides whether an answer recommends anyone far more than the engine does. Our own measurement, August 2026, with the stance of every mention classified by model and the mentions themselves matched deterministically.
The mechanism
What we do with one call.
Every call runs the same three passes, one call at a time, against the Foundation you approved.
-
1
Triage
We read the call and decide whether it counts. Calls where you are not the offering under evaluation are dropped before anything is extracted, and only customer and prospect voice counts toward a theme. A messy archive is the normal input.
Our job, not yours -
2
Extract
We pull what your buyer actually said: the words they use for the problem, what they pushed back on, what put them in the market, and who they were weighing you against. Every snippet keeps its speaker role, its call, and its date.
Verbatim, attributed -
3
Reconcile
We count each row of the map against the calls. Rows your buyers confirmed get weight. Rows none of them ever echoed get flagged. Themes the map never had get proposed as additions, and you decide.
Against your Foundation
Nothing is inferred and presented as evidence. A theme that no buyer said out loud does not become one your report claims they did.
What changes
Same crawl. Sharper questions.
Both question sets are complete, valid, and 150 questions long. What changes is what they are made of, and two categories of question exist only when buyers have spoken.
- Their words replace our inference. The phrases your buyers use for the problem, before your marketing renames it, become the phrasing we put to the engines.
- Every objection becomes a question. What they pushed back on, and what nothing on your site answers yet, gets measured. Without calls, no objection questions exist at all.
- Every buying trigger becomes a question. The moment a purchase actually starts, in their account and their words. Without calls, no trigger questions exist at all.
- Confirmed rows get more of the 150. What your buyers validated outranks what a rating inferred, so the set spends its budget where the evidence is.
- Your set says which mode produced it. Every question set is stamped buyer-validated or public-evidence-only, so you always know which one you are reading.
Then the plan that comes back answers objections you can cite by call and date, which is a different conversation with your team than a plan built from a guess.
What we need
Transcripts, and whatever you have.
A summary is a model's account of the call. The transcript is your buyer's own words, and the words are the part we can put in front of the engines. Export from Fathom, Gong, Gemini, Granola, Fireflies, Zoom, or anything that writes a transcript to a file, and drop them in your account. We take .txt, .vtt, .md, .csv, .json, and .docx.
There is no minimum. We would rather read a messy archive than wait for a curated one, because the triage pass exists to do that sorting for you. Your transcripts live in your account, you add and remove them yourself, and a brand shared with a colleague is read-only to them.
Every cycle reads whatever you recorded since the last one, so the questions keep moving toward what your buyers actually raise.
If you don't record calls
The crawl still runs.
Your 150 buyer questions are written from the Foundation you approved, scored across ChatGPT, Claude, Gemini, and Perplexity, and the Visibility Report says plainly which rows rest on public evidence only. You get the report and the ranked action plan on the same terms, charged on delivery.
Start recording now. Anything recorded before your next cycle is read into the next Foundation, and the questions move with it.
Frequently asked questions
Do I have to share sales calls?
No. The crawl runs either way. Without calls we write your 150 buyer questions from the Foundation you approved and the report says plainly which rows rest on public evidence only. Sharing calls changes what the questions are made of, not whether you get them.
Transcripts or summaries, and how many?
Transcripts, and whatever you have. A summary is a model's account of the call. The transcript is your buyer's own words, and the words are the part we can put in front of the engines. Export from Fathom, Gong, Gemini, Granola, Fireflies, Zoom, or anything that writes a transcript to a file, and drop them in your account. We take .txt, .vtt, .md, .csv, .json, and .docx.
What if most of our calls are not useful?
That is the normal input, and sorting them is our job. The pipeline triages every call before it extracts anything, drops the ones where you are not the offering under evaluation, and counts only customer and prospect voice toward a theme.
What does it actually change about the questions?
Two whole categories of question exist only when buyers have spoken. Every objection your buyers raised becomes a question we put to the engines, and every buying trigger they named becomes another. Your buyers' own phrasing replaces our inference, and the rows they confirmed get more of the 150. Your question set is stamped with which of the two modes produced it.
Who reads the calls?
The extraction runs one call at a time against your approved Foundation, and a person validates the output before it reaches you, the same way a person validates every engine answer in your report.
Is Voice of Customer a separate purchase?
No. It is part of the $997 AI Visibility Crawl, alongside the crawl itself, the Visibility Report, and the ranked action plan. The Foundation and your 150 questions stay $0.
No card required
Start with the map. Then let your buyers correct it.
Your Foundation is free, typically in your account within an hour, and revised with you until you would stand behind it. Share your calls when the map is right, and we sharpen the questions before we measure a thing.