River
Y CombinatorBacked by Y Combinator
FREE TEMPLATE

Messaging Framework From Customer Language

Four documents and three sheets that count your customers' words against your company's, then put a numbered verbatim behind every line.

Free download  ·  No account needed

Language Frequency Analysis

Their word against ours, counted

46 sources  ·  30 sales calls, 8 interviews, 6 months of tickets, 8 public reviews  ·  Larkfield, inventory software

ConceptWhat customers sayUsesWhat we sayTheir uses
The core problemflying blind on stock31real-time inventory visibility0
What good looks likeknow what we actually have27single source of truth0
The blocking constraintwe are not ripping out the ERP14integration-friendly architecture0
The finance frictionfinance does not believe our number9auditability1
The loss eventshrinkage7inventory variance2
Improvementno customer term found0optimise1

The line that ends the argument

Customers said “single source of truth” zero times across forty-six sources. The messaging document it replaced used it four times, and the homepage still uses it twice.

Two verdicts nobody expects

Keep the jargon. “Shrinkage” survives at seven uses because customers genuinely say it. Replacing it with plain language would be an error the count already ruled out.

Make no claim. Speed to value has three mentions, two of them negative. The framework says nothing about it, and says so in writing.

Every count is split by source type and marked for whether it was unprompted, because in a sales call the rep speaks first.

Larkfield's messaging document says "unified inventory intelligence platform" and "single source of truth". Across forty-six sources, thirty sales calls, eight interviews, six months of support tickets and eight public reviews, customers said "single source of truth" zero times. They said "we are flying blind" or a close variant thirty-one times, nine of them using that exact phrase. Both wordings were available to whoever wrote the document. Only one of them came from a transcript.

This is why messaging arguments do not resolve. Two people argue from taste, taste has no tiebreaker, and the more senior preference wins. A count has a tiebreaker. Zero uses in forty-six sources against thirty-one is not a matter of opinion, and nobody disputes it without reading the transcripts, which nobody does. So the frequency sheet is filled before the framework, because a framework written first gets defended with evidence found afterwards.

Then the words go somewhere public, and a customer quote on a page is an endorsement under 16 CFR Part 255. An excerpt has to fairly reflect what was said, and a user testimonial rests on somebody who still is one. So every verbatim here carries its full surrounding passage rather than the pull quote, its consent scope by surface, and the date its user status was last confirmed. It installs into the marketing workspace beside the Search Console keyword map and the rest of the template library.

Every claim traced, every quote cleared, and the competition ranked honestly

The Claim to Evidence Map with statuses, the Proof Point Register with consent and confirmation dates, and the alternatives ranked by mention count.

Claim to Evidence Map

One row per claim per surface. Four statuses, and one of them takes copy off a live page.

ClaimWhere it is liveSourcesTypesStatusGap
You promise stock you cannot shipHomepage hero; ads314cleared
It reads and writes to what is already thereHomepage; every deck142cleared
Every adjustment carries who and whyProduct page; finance one-pager42cleared
Finance stops arguing with operationsCase study headline11blockedThe one passage carries a qualification that cannot be trimmed away
Built for wholesale rather than retailBattlecard31needs workOne source repeats a phrase the rep used earlier in the same call
Reduces the year-end write-downFinance one-pager22needs workBoth sources call it a surprise. Neither says it got smaller
Cuts carrying cost by 20%Sales deck slide 600removeNobody in the building can name where the 20% came from
Live in weeksDeck slide 9; pricing page00removeThree mentions in the corpus, two of them negative
Easier to use than the alternativesHomepage subhead00removeFour passing mentions, and every competitor claims it

Three claims come off live surfaces. The 20% is the one that matters: it is on slide six of every deck, it has been quoted in a board meeting, and tracing it back finds no source in the corpus and nobody who can name its origin.

Blocked is not the same as removed. The finance claim has real evidence and cannot be used as written, because the only passage supporting it carries a qualification. Quote the qualification too, or paraphrase without quotation marks.

Two source types beats four sources of one type. Four verbatims from four sales calls all share the same contamination risk.

Proof Point Register

Full passage, not the pull quote. Consent scoped by surface. User status with a confirmation date.

IDSourceUsing at the timeUnpromptedConsent scopeConfirmedPublic use
V-029InterviewCustomer 5 yearsyesSite, deck, paidquarterlycleared, all surfaces
V-011InterviewCustomer 14 monthsyesCase study, site, deckquarterlycleared, full passage required
V-033InterviewCustomer 3 yearsyesOwned surfaces onlyquarterlyowned only, paid needs re-consent
V-019Public reviewCustomer 2 yearsyesExcerpt with attributionoverdueconditional: passage qualifies the praise
V-004Sales calltrial at the timeyesNone soughtn/ainternal only, not a bona fide user
V-041Support ticketCustomer 18 monthsyesNone soughtoverdueinternal only, speaker did not know
V-037Public reviewchurned at month 16yesNonelapseddo not use
V-035Sales callProspectnoNone soughtn/anot evidence: rep said it first

V-037 is the row this register exists for. A glowing public review from somebody who churned four months later. Under 16 CFR 255.1(c) a user endorsement needs a bona fide user at the time plus good reason to believe they remain one, and nothing in a normal marketing process was ever going to catch this.

V-041 is a support ticket, and it is the best-worded quote in the corpus. It stays internal. The person writing it did not know they were being read for marketing and no consent exists.

Every row holds the passage around the excerpt. A quote ending just before the word “but” is the case that field exists to catch.

The alternatives, ranked by mention count

What customers said they were actually choosing between, in the order the corpus put them.

AlternativeMentionsWhat they said about it
Do nothing and count harder19“the plan right now is we count more often, that is the plan”
A spreadsheet somebody maintains16“Dev built it, Dev left, nobody knows what the third tab does”
The ERP module they already own12“we already paid for it, nobody could get it working”
Named competitor A7“quoted us more than the problem costs”
Named competitor B3“built for retail, we are wholesale”

Doing nothing outranks both named competitors combined, roughly two to one. The three largest alternatives are not vendors at all, which means the competitive deck is aimed at a fraction of the real competition and the framework should lead on the cost of the status quo.

The third row is a competitor already inside the account. Twelve customers are paying for inventory functionality they could not get working, which is a different sales conversation from a head-to-head and nobody was having it.

This ranking is the slide that changes behaviour, and it takes one afternoon of counting to produce.

What's in the pack

01

Language Frequency Analysis

Their term against yours with counts split by source type, marked for whether each use was unprompted, and a verdict per concept.

02

Claim to Evidence Map

One row per claim per surface, with a status that includes remove, so unsourced copy comes off live pages rather than staying.

03

Proof Point Register

Every verbatim with its full surrounding passage, consent scoped by surface, and the date the speaker's user status was last confirmed.

04

Messaging Framework

Claims ordered by count rather than by preference, each carrying verbatim numbers, with the old wording and its zero beside it.

05

Voice of Customer Appendix

The passages at length, grouped by what they evidence, plus a written note on every claim the corpus could not support.

06

Verbatim Usage Standard

The four checks before a quote goes public, built on what the FTC's endorsement guides actually require.

07

Rollout Note

What to present in what order, what each team changes on Monday, and the answers to the four objections you will get. Route anything sales keeps hearing into the objection library instead of drafting a one-off answer here. Before the rollout meeting, run the headline claim through a message test rather than trusting whichever version the room prefers.

How to use it

  1. 1

    Open in River, or take it blank

    Open the pack in River and send your transcripts, or download the Word documents and CSV sheets and count by hand.

  2. 2

    Send whatever you have

    Sales calls, interviews, a ticket export, your own public reviews, lost-deal notes. Six transcripts produce a real finding.

  3. 3

    Read the widest gap

    Two numbers: how often customers used their phrase, and how often they used yours. That comparison is the whole argument.

  4. 4

    Trace what is already live

    Every claim on your homepage and in your deck gets a status. Most teams find one nobody can source and one quote from a churned customer.

Frequently asked questions

Is this template free?

Yes, with no account, no card and no email gate on the download. Edit with AI is the optional half: River reads your transcripts, runs the counts, and fills the framework with verbatim numbers behind each line. Every pack sits in the template library.

How many transcripts do I need?

Six sales calls produce a finding worth acting on. Forty produce one nobody argues with. The document you are replacing was almost certainly built on none, so the comparison is not against perfect research, it is against a workshop.

What if I have no call recordings at all?

Your corpus exists anyway. Public reviews of your own product are unprompted and quotable. Support tickets are the most honest source you have. Lost-deal notes count too, and re-reading those rather than trusting the loss reason field is where the positioning pack starts. Begin with whichever you can reach today.

Why keep jargon customers use?

Because the test is the count, not the plainness. The worked example keeps "shrinkage" at seven uses because customers genuinely say it. Replacing a customer's own word with a simpler one you invented is the same error as the original, in the other direction.

Why does a template care about FTC guides?

Because the point of this pack is putting customer words in public, and a quote on a page is an endorsement. The guides require that an excerpt fairly reflect what was said and that a user endorsement rest on somebody who was, and still is, a real user.

Won't this just produce dumbed-down copy?

It produces longer copy, more often than shorter. The worked example's problem statement is twice the length of the phrase it replaced, because a sentence the reader recognises as their own situation beats a compressed one they have to translate. Take the counted language into a homepage rewrite with a claim-to-proof map next.

What does Edit with AI actually do?

It signs you up, installs this pack as a private workspace, and starts counting. Send the transcripts and your current messaging document, and the widest gap between the two comes back with a full verbatim passage beside it.

Find out how often customers say your headline

Take the Word documents and CSV sheets blank, or open this exact pack in River and send it six sales call transcripts.

Edit with AI