the problem · 1 of 4 — the moment
Not a search — a question. And it isn't only buyers: prospects, partners, investors and future hires open the same way.
the problem · 2 of 4 — the shortlist
Pulled from the sources the engines already trust — and the same three names come back whether it's a buyer, a partner or an investor asking.
the problem · 3 of 4 — the miss
Ask again tomorrow and the order shuffles — always among the same sourced names. Miss the sources and you miss every version of the answer, to everyone who asks.
the problem · 4 of 4 — the silence
No click, no impression, no lost-deal report — just prospects, partners and investors quietly handed someone else's name. The only way to know is to ask the engines, many times, and count. That's what Tallyvox does.
Why the results differ
The industry’s to-do list is mostly right. The execution mostly isn’t — and the hardest items rarely get attempted at all. Here’s the full ledger, and how we ship it:
Why the source work leads: across 75,000 brands, branded web mentions are the strongest correlate of AI visibility
(Ahrefs, ρ=0.664 — correlational, and we say so). The count is how you see it working on YOUR market.
You run your own playbook — you need the honest scoreboard: who the engines name, how often, receipts stored, confidence intervals on every number. Take the count and run.
Same count, plus the seven fixes worked for you — flagship citation outreach first, because that’s where the evidence points. Every month ends with the same yardstick, so you see exactly what the work did.
The receipts
Ask an AI engine the same question a hundred times and you’ll get the same list back fewer than once — measured across 2,961 collected responses, not rhetorical. So we run every prompt repeatedly, store every raw answer verbatim, and report frequencies with confidence intervals. Flip through what stored answers look like.
Because this industry sells screenshots. We store the raw response for every run of every prompt on every engine — so when your report says “you appear in 12% of answers,” you can open the 12% and read them. And the 88%.
Numbers you can audit are numbers you can act on.
Every month we ask ChatGPT, Claude, Perplexity & Gemini the questions your buyers ask — and count who gets named. Counted, not claimed. Is it you?
The Tallyvox Ledger
Data stories, methods and playbooks from measuring who the AI engines actually name — everything sourced, nothing padded.
How it works
We read your site and build ~40 of the real questions people ask AI when they’re looking for a company like yours — the way buyers, partners, investors and job-seekers actually phrase it — then run each one again and again across six AI engines, because a single answer is noise. Every raw response is stored. You get a visibility score with a confidence interval, not a screenshot.
AI answers are assembled from sources — review sites, comparison lists, directories, communities. The scan produces the exact list of domains the engines cite in your market, ranked by how often, marked where you’re missing. That list is the roadmap.
Listings claimed, real reviews generated, editors pitched, answer-shaped pages published — then the identical prompt set is re-measured monthly. Movement (or its absence) is visible, attributable, honest. Progress you can count, month over month.
What you get
A measurement company should be able to itemise its own deliverable. So here’s the manifest for the full audit — exactly what lands in your inbox, about a week after kickoff.
The free snapshot ships items 1, 2 (top three only) and 5 at teaser depth — enough to know whether the problem is real. The retainer re-ships this manifest monthly, plus the work that changes it. See a sample report (PDF, illustrative data) →
figures shown on this website are illustrative; your documents use your market’s stored answers
Pricing — on the page, like adults
Hiding prices behind a form costs everyone a week of email. Here they are. All measurement, no lock-in — the retainer is month-to-month after an initial three.
The Snapshot — find out if you have a problem, free, no call required.
Free scans are paused. We reopen them in batches.
The Full Audit — the complete measurement and the plan. Yours to run with, in-house or with us.
The Retainer — we work the gaps and re-measure monthly with the identical yardstick.
The audit credits 50% toward month one of any retainer started within 30 days.
Flat pricing, honest scoreboard. You pay for the work and the count — and the count shows you exactly what the work did.
MSP math: one retained client is typically worth $1,200–1,500/mo to you. The count paying for itself requires one.
Take this shopping
Plenty of vendors would rather you didn’t ask these, so here’s the checklist we’d use to buy what we sell. Ask these of everyone — click each for our answer, on the record.
Yes — repeatedly, across six engines, reported as frequencies with 95% confidence intervals. A vendor showing one screenshot is showing you a coin flip.
tallyvox: passYes — every run stored verbatim. Pick any percentage in the report and we’ll open the answers it came from.
tallyvox: passWe promise the lever with the strongest evidence behind it — presence in the sources engines actually cite, worked every month — and a recount on the same yardstick that shows exactly what it moved. A number you can audit beats a number you were promised.
tallyvox: pass — receipts, not hypeNot the whole job — ever. We do both as part of crawl hygiene, done right: llms.txt kept current and aimed at answer-shaped pages, schema kept clean where it earns classic-search real estate. But the lever that correlates with inclusion is presence in the sources engines actually cite — that’s the flagship work, and the count proves whether any of it moved.
tallyvox: pass — hygiene plus the flagshipYes — it’s the audit’s headline section: every cited domain, ranked by frequency, marked where you’re absent, with a route in for each.
tallyvox: passYes — with the confidence intervals that prove it. Wins only count when a scan sees them; retainers go month-to-month after the first three so honesty stays cheap for you and expensive for us.
tallyvox: passchecked 0 of 6 — click through them all — every answer is on the record
Tallyvox started with an engine: it reads a company’s site, builds the real questions people ask AI when they’re looking for a firm like it, asks the engines over and over, and turns the noise into honest frequencies — reports where every number traces to a stored answer.The service grew around that engine — finding the sources AI trusts in each market, and doing the placement work that gets clients counted.
As the team grows, one thing doesn’t change: everything stays auditable. Whoever researched your market, pitched your listings, or pressed the scan button — the receipts are the same, and they’re yours to inspect.
built in-house · every answer stored · counted monthly · meet the team →Every accent on this page is a choice. Pick the one that moves you — the whole site puts it on, just for you, just on this device.
Choose your colour →referral program
Send businesses our way and earn 5% to 40% of everything they pay — audits and retainers alike. Your link credits you for 90 days from the first click.
25 minutes with a human who’ll show you real scans — no deck, no pressure. If the count says you’re fine, we’ll tell you that too.