Your positioning is being graded without you
Someone audits your product marketing every day, and it is not a consultant. Each time a buyer asks an AI assistant what your product is, which vendors to consider, or how you compare to the competitor they already know, a machine reads your homepage, your documentation, your comparison pages, and your reviews, then returns a verdict in a confident neutral voice. 73% of B2B buyers now use AI tools like ChatGPT and Perplexity in their purchase research, and the buyers who arrive from those answers convert at 14.2%, roughly five times the rate of Google organic traffic — the machine has already done the qualifying. That verdict reaches your buyers today, and nobody on your team has read it.
This is the opportunity, not the threat. Product marketing has never had an accepted measure of its own impact — sales has quota, demand generation has pipeline, product has adoption, and each of those arguments settles itself. Product marketing gets judged on the launch calendar, because that is the only artifact anyone can count. That single absence explains why the function gets briefed late, staffed as a service desk, and asked to defend its value in a medium no serious executive believes.
What has changed is that the quality of positioning now produces a checkable output. You can read what the market thinks you are. The mechanics of how those answers get assembled are the subject of the companion piece on AI search; this one is about what to do with what you find. It covers the ten-minute check, the four links that hold positioning together, and the charter that falls out of running it twice — and it is written to be forwarded, to your team or to your CEO.
Read your own verdict first
Before any method, do this. It costs ten minutes, needs no budget or approvals, and it is the most useful thing in this article. Open a fresh session in ChatGPT, Claude, Perplexity, and Gemini — no history, no account context — and ask four questions:
what is [your product] · best [your category] tools for [your buyer] · [your product] vs [your main competitor] · what are the drawbacks of [your product]
Paste the answers into one document and read them as a stranger would. Three things to look for, in order of how much they should worry you.
Does the engine use your category word, or a different one? If your homepage says one thing and every engine says another, that gap is costing you the shortlist.
Do you appear on the unqualified category question, or only when the question already contains your niche? Appearing only in the narrow version means the market has filed you as a specialist in a corner of the category, whatever the homepage claims.
What does the engine volunteer that your pages never mention? Limitations, pricing, caveats, a competitor's characterization of you. This one matters most, because whatever the machine adds is precisely what your own comparison content failed to say first — and it is being served to your buyers now.
Most people who run this find something they did not know. What follows is the method for fixing what you find.
The proof chain: category, source, owner, mechanism
Positioning that cannot be checked is not positioning, it is preference. The proof chain is what turns one into the other, and it has four links — four words, in order, that a marketing leader can hold in their head and repeat to a CEO.
Category. Use the word the market already uses back to you, not the one invented in an offsite. A category you have to teach is a tax on every deal.
Source. No number ships without something a reader could look up.
Owner. Every gap names the function that can close it, which is frequently not product marketing.
Mechanism. If a number moved, something has to say why. A holdout, an adjusted trend line, or a written estimate you would defend.
They run in that order and the order is load-bearing. Sourcing claims for a category nobody recognizes is decoration. Measuring outcomes before either is arithmetic on sand.
| What used to count as proof | What counts now |
|---|---|
| The launch shipped on time | The market can state what you do, in your words |
| The messaging doc says we are differentiated | A stranger can check the differentiator without calling you |
| Our claims are compelling | Every claim carries a source someone could look up |
| Sales says the messaging works | The words reps use are measured against the words you approved |
| We have a battlecard | The comparison a buyer reads is one you wrote, not one your competitor wrote about you |
| Win rate rose after the campaign | A holdout or a disclosed estimate says the campaign is why |
Every line in the right-hand column is the same claim you were already making, with one addition: someone outside the building can verify it.
Lead with the word the market already uses
The most expensive positioning mistake is not a weak differentiator. It is leading with a category label nobody says back to you, because every downstream asset inherits the confusion and every engine files you somewhere you did not choose.
The pattern repeats. A company decides it has outgrown its category — expense software becomes a spend management platform, a ticketing tool becomes a service intelligence suite — and the homepage changes on Monday. The analysts do not change. The review sites do not change. The engines, which learn from both, do not change. So the brand runs two categories at once: the aspirational one it announces and the operational one the market keeps using. Buyers searching the real category find competitors. Buyers searching the aspirational one find nothing, because it has no search volume yet.
The obvious question is how you find your real word, and it takes about an hour. Read the category label on your own G2 or Capterra listing. Read the market name in the URL of your analyst profile. Read what the four engines called you in the ten-minute test. Read the words in the comparison content competitors write about you. Where those agree, that is your word, whatever the homepage says.
Then the rule to hand your team: you do not get to name your category, you get to earn it. Lead with the word you have been given, put the ambition one line beneath it, and treat analyst ratification as the milestone that lets you promote it. Coining a category is a legitimate strategy and occasionally a brilliant one — but it is a multi-year campaign to change what other people say, not a copy change, and running the new word before anyone else says it costs you the shortlist in the meantime.
Anchor the numbers you are loudest about
Across the audits I have run — a handful, all B2B software — the statistics with the most prominence consistently had the least evidence. Homepage stat tiles, in the largest type, under a header promising measurable results, carrying no study, no customer, no sample, no method. The claims buried three levels down were meticulously sourced.
This is not only a credibility problem. The Princeton and Allen Institute research on generative engines measured which content features actually get picked up in AI answers: quotations lift visibility 27.8%, statistics 25.9%, and cited sources 24.9%. Machines quote what is quotable, and an unsourced number is not. What makes this diagnostic rather than sloppy is the contrast beside it. The same companies cite major research firms by name, precisely, for market claims about their category. The discipline exists in the building. It simply never gets pointed at claims about their own product's effect, which are the only claims a buyer needs verified. Somewhere along the way, sourcing became something you do to prove the problem is real, not to prove your solution works.
The fix is a list, not a philosophy. Put every load-bearing number in one place, mark each anchored, customer-attributed, unsourced, contradicted, or stale, and work from the top of the page down. It takes an afternoon and it usually finds something alarming — in one audit, a single customer outcome appeared as three different percentages on three pages. Another surfaced a category of error I now check for every time: "certified" claimed against standards that issue attestations, or that have no certification scheme at all.
The anchored version is longer, and that is the point: specificity is what makes a claim quotable — by a buyer building an internal case, by an analyst writing a profile, by a machine assembling an answer. Vague superlatives get dropped in synthesis because there is nothing to carry. If you cannot source a number, cut it: an unanchored claim is not neutral, it is a liability the first skeptical reader converts into doubt about everything near it.
Name the owner, or the finding dies in the meeting
Two things reliably kill an honest audit. A finding that lands on product marketing when product marketing cannot fix it, and a real gap softened into vagueness to protect a colleague. Both are solved by the same move: state the problem at full strength, and name the function that owns the fix separately.
Most of what looks like a messaging failure is a distribution failure with a different owner. Product marketing writes the message; whether the field holds to it is coaching and enablement. Product marketing writes the asset; whether a rep finds it at the moment of need is operations and tooling. That distinction is the whole argument of Kill the Battle Card.
One audit found a company whose competitive content was genuinely better than its rival's — sixteen comparison pages that conceded real competitor strengths, where the rival's conceded nothing. It lost the comparison anyway. The pages sat in a blog subdirectory with no sitemap entry and no navigation path, so the rival's weaker page ranked first, and the answer engine adopted the rival's characterization of them as its own neutral description. That is not a writing problem. Nobody needed to write anything. It needed a sitemap entry.
Getting this right changes what the audit is politically. A report that grades product marketing on levers it does not hold is a report the function never commissions twice. A report that says here is the gap, here is its business consequence, here is the team that can close it is something a marketing leader can hand a peer without it reading as blame, and something a CEO can act on, because it arrives as an operating plan with names on it.
Say why the number moved, or do not claim it
The last link buys credibility for the other three, and it is the one marketing skips most often. If win rate improved after the repositioning, something other than the calendar has to connect them. Otherwise you are asking a board to accept correlation from the one function famous for presenting it.
Three mechanisms, all cheaper than they used to be. A holdout: ship the new message to part of the sales org, keep a comparison group on the old one, compare after a quarter. This was once politically impossible and logistically expensive; with AI drafting the variant assets and CRM automation splitting the cohorts, thirty days is a realistic ask. A trend line: the before-and-after trajectory, adjusted for confounds you can name — seasonality, segment mix, a pricing change. A disclosed estimate: your honest judgment of how much of the movement the work is responsible for, with the assumptions written down where someone can argue with them.
That third one deserves defending, because it sounds like a cop-out and is not. An estimate whose method is written down and challengeable is a scientific object. A confident number with no method behind it is a coincidence wearing a suit. The difference between them is the difference between a marketing leader a CFO trusts and one they humor.
What one looks like written down — invented, to show the shape of an honest write-up, not a real program or result.
Two-thirds of the enterprise segment moves to the new comparison narrative on 1 March; one-third stays on the old battlecard. Ninety days later the treated cohort closes at 31% against the control's 24%, with average deal size flat in both. The write-up reads: seven points, one quarter, one segment, not seasonally adjusted, 140 deals — we would defend about half of it, and here is why. That last clause is what makes it credible. A number nobody has argued with is not evidence, it is decoration.
What makes this work is refusing credit you cannot support. When the audit finds outcomes moving with no mechanism behind them, it says so and discounts the claim — which feels like losing until the first time you present it and a board realizes it is reading a marketing document that argues against itself.
The charter that falls out of it
Run it twice and what you have is a charter, which is the thing the function has actually been missing. It states the job — accountable for whether the market can tell what you are, why you win, and what it is worth. It draws a boundary, protecting the team from being graded on levers it cannot pull and removes the excuse for everything inside it. And it gives everyone a number that is not a launch count: the movement between this quarter's picture and the next.
If you sit in the CEO seat, run it in reverse. Do not ask whether product marketing is working — every sincere answer to that question is a narrative. Ask what the market says you are, which claims on the homepage carry a source, and what would have to be true for the last launch's numbers to be causal. The request takes one sentence and the first honest answer takes a month.
Thirty days
Days 1–10: build the picture. List every category label you use and where. List every load-bearing claim and mark which carry a source. List your differentiators and run each through the swap test — put a competitor's name on the claim, and if it still reads true, it is category advertising you are funding alone.
Days 11–20: fix in order. Contradictions first, because they cost trust immediately. Then the loudest unsourced number. Then the category word, everywhere at once — homepage, About block, analyst profiles, review-site listings, on the same day.
Days 21–30: instrument one thing. Pick a single in-flight initiative and give it a mechanism before it ships. One holdout, honestly run, is worth more than a year of confident correlation.
What comes back is a picture, not a grade — every category label you use beside the one the market uses, every load-bearing claim marked anchored or unanchored, every gap tagged with its owner. It looks like this:
| Link | What it found | Owner |
|---|---|---|
| Category | Nineteen distinct labels in use across eight pages; every analyst and engine uses one word, and it is not the homepage's | Product marketing |
| Source | The six largest homepage statistics carry no study, customer, sample, or method; the market claims beside them cite three research firms by name | Product marketing |
| Owner | Sixteen strong comparison pages, none in the sitemap and none reachable from navigation | Web and SEO |
| Mechanism | Win rate up and cycle time down since the repositioning, with nothing separating the work from the quarter | Revenue operations |
The Positioning Register
The worksheet version of the four links, with the four engine questions built in at the top. Built to be forwarded to your team with two words: fill this.
Have it run with you
Paste your pages, your comparison asset, and what the engines said. It builds the picture, refuses to judge anything it cannot see, names the owner on every gap, and hands back the fix list in priority order.
Look before they do
The alternative to this audit is not going unaudited. It is being audited continuously, by machines, in front of your buyers, with no chance to see what they found. Ten minutes and four questions is the entire cost of looking first.
Underneath the method is a claim about the craft I would defend anywhere: positioning that cannot be checked is not positioning, it is preference. Category, source, owner, mechanism are what turn a point of view into something a stranger can verify, a colleague can act on, and a board can fund. That is also, not coincidentally, what makes a brand the one the machine quotes.
I have run this, not only written it. The findings in this piece — the nineteen labels, the orphaned comparison library, the certifications claimed against standards that do not issue them — came out of real outside-in audits of real B2B software companies, and the anonymity is theirs, not a hedge. This site is built to the same standard: every external statistic here links to its source, every illustrative example is labelled invented, and anything I cannot evidence is not published. View source.
The worksheet and the skill are free, like everything else I publish. If you want this built into how your team works rather than just read about — reach me at daniel@cmoconfessions.com, or leave your email below. No gate, no sequence.
← Home AI search Loop design Kill the Battle Card The Skill Stack