The Claim File

Positioning is one versioned file in the repo that every agent, page, and assistant profile loads. Sort every change by the claim, and track story lag.

Falk Gottlob8 min readNew

Kristen Bryant Smith at Appcues surveyed 60 marketers with Navattic in October. 75% said their product ships faster than last year. 7% said their launch process keeps up.

A faster launch process won't close that gap, because the launch is the wrong unit. The product changes every day now, and the first reader of what you say about it is usually a model. So positioning has to live where the product lives, as a file the build can read.

That's the claim file. It's the product marketing seat's core artifact in The Product OS, the same way the eval is product's.

The short version

  • The claim file is one versioned file in the repo that holds the positioning and its proof. Every customer-facing agent loads it, every proof surface is checked against it, and every assistant listing is generated from it.
  • One question at merge time keeps it current: does this change the claim, the proof, or neither? Two numbers say whether it's working.
  • Story lag is the days from a change shipping to the assistants describing it correctly.
  • Message drift is the share of answers that contradict or miss the current claim. Product marketing writes the file and owns the claim.
  • The CPO owns the pipeline that carries it, which is why the seat sits in the product org.

What goes in the file

April Dunford said it at Mind the Product in July: "The fundamentals hold. The execution doesn't." So the file starts with the fundamentals. Who it's for. What they'd do instead. Why you win.

Then the parts a model needs and a slide never had. The proof for each claim, with a link and a date. The words you use and the ones you retired. And the never-say list: stale pricing, the old product name, the capability you sunset.

Melissa Kwan built a version of this at eWebinar and described it in Kyle Poyar's Growth Unhinged. Customer quotes, objections, alternatives, a language guide, and instructions the model maintains, with every finished page going back in. Her line is the one to keep: "positioning is retraining, and retraining only works if every touchpoint says the same thing."

The file sits next to the spec and has two owners, the PM and the PMM. Part 2 of The Product Marketing Transition makes that case, so I won't repeat it here.

Claim, proof, or neither

Every change gets one question when it merges. Does it change the claim, the proof, or neither?

Neither is most of them. A faster startup, a new flag, a fixed edge case. An agent drafts the changelog entry from the diff and nobody meets about it.

Some changes move the proof. A new integration, a benchmark, a capability a competitor lacks. The claim stands, but the evidence changed, so the docs, the comparison pages, the pricing page, and the assistant listings move the same day.

A few move the claim itself. A new buyer, a new alternative you now beat. Those get reviewed like a spec change, by both owners, before the merge.

This replaces the launch tier. Tiers rank a change by the size of the feature. Buyers and assistants care about what it does to the story, and part 3 has the longer argument. The campaign calendar stays with marketing. The claim, and every change that touches it, is product's.

Every reader loads it

Salesforce reported on November 3 that half of Slack messages now come from apps and agents. Your messaging has the same mix of readers.

Your own agents come first, because you control what they load. Sales, support, onboarding. Each one loads the claim file as context, and none of them carries its own copy of the pitch. A copy drifts the day after you paste it.

The buyer's assistant is the reader you don't control. Louise Linehan's Ahrefs analysis of 260 million prompts across 10,000 brands found that Reddit mentions correlated most with how often an assistant names a brand, at 0.75. That's a correlation, and Ahrefs says so. But it tells you what the file needs for this reader: proof that other people can repeat in their own words.

Since November 10, that second reader has a front door. OpenAI's Businesses in ChatGPT beta, as Luis Rijo reports at PPC Land, lets a company claim a profile, add a product feed, publish a Business Agent, and see mention and click counts. Read as a product leader, that's three more surfaces a release has to update. The feed comes from the catalog. The agent has instructions and a test set. The profile is the claim, compressed. Part 9 of The CPO Transition covers who decides the terms. This chapter's rule is simpler: all three are outputs of the file, never hand-edited copies of it.

People still decide. More and more they read what the first two readers told them.

Two numbers

A messaging doc nobody tests is a guess. The file gets an eval suite, the same as any other model input.

Write down the questions buyers ask before a shortlist, twenty to thirty of them. For each one, write a rubric from the file. Did the answer name the alternative you beat? Did it give the proof? Did it avoid the never-say list?

Run it inside on every change to the file, against your own agents, the way a prompt change runs against product evals in Prompt Ops. Run it outside weekly, against the public assistants.

Two numbers come out.

Story lag is the days between a change shipping and the assistants describing it correctly. It's the story stage's measure in the Impact Loop. The surface it measures moves fast. Rich Holmes reported Ahrefs data on about 37,500 AI citations where at least half were gone by day two, and about 75% of those came back later.

Message drift is the share of answers, inside and outside, that contradict or miss the current claim. Inside drift you can fix in an afternoon, because it's a context problem. Outside drift tells you where the proof is missing.

Part 4 walks through the suite in more detail.

Why this sits in the product org

The merge question is asked by people looking at a diff. The agents that load the file have prompts in the repo and evals at the gate. Assistant listings update when a release does. All of that is the product pipeline.

Put the file's owner in another org and the review becomes a ticket. The change merges on Tuesday, the ticket gets read on Thursday, and the assistants have told the old story since Wednesday.

So product marketing writes the file and owns the claim, as a seat in the product org. Design holds the same position with the design system agents generate from. Two context files, one pipeline, one owner of the outcome. That's the practical reason The CPO Transition keeps landing on one function.

What I expect

These are my predictions, not facts.

By the end of 2027, at AI-native software companies, every customer-facing agent loads one versioned claim file, and a change to it has to pass a message eval before it merges.

Within 12 months, "update the assistant listings" is a line on the release checklist, generated from the same file as the docs.

By 2028, story lag and message drift sit on the CPO's weekly page next to the repair rate, and the head of product marketing reports both.

One thing to try this week

Find every place your positioning is pasted today. The sales agent's prompt, the support agent's prompt, the website, the pricing page, the deck, any assistant profile you've claimed. Count the copies and note the date each one last changed.

Then write the one file, load it into one agent, and delete that agent's pasted copy. Ask it the ten questions your buyers ask before a shortlist. That's your first eval run.

Sources: Kristen Bryant Smith, The State of Product Launches, Appcues with Navattic, 2026-10-27 · April Dunford at Mind the Product London, 2026 · Melissa Kwan in Kyle Poyar's Growth Unhinged, 2026-10-21 · Salesforce, Slack stats, 2026-11-03 · Louise Linehan, Ahrefs, 2026-11-03 · Luis Rijo, PPC Land, on Businesses in ChatGPT · Rich Holmes, Department of Product, 2026-11-01

Part of The Product OS, in the Product marketing seat. The running argument is in The Product Marketing Transition. Related answer: Half of Slack messages come from agents. Who writes the messaging they read?

Share this post

Frequently asked

What is a claim file?+

One versioned file in the repo, next to the spec and the agent prompts, that holds the positioning: who the product is for, what they'd do instead, why you win, the proof for each claim, and the lines you never want said. Every customer-facing agent, proof page, and assistant profile is generated from it or checked against it. The PM and the PMM own it together.

What replaces launch tiers when the product ships daily?+

One question at merge time: does this change the claim, the proof, or neither? Neither is most changes, and they flow from the pull request to the changelog and docs with no meeting. A proof change moves the docs, comparison pages, pricing page, and assistant listings the same day. A claim change is reviewed like a spec change by the file's two owners before it merges.

What are story lag and message drift?+

Story lag is the days between a change shipping and the AI assistants describing it correctly. Message drift is the share of answers, from your own agents and the public assistants, that contradict or miss the current claim. Story lag is the loop's measure for the story stage. Drift tells you which reader is behind.

Who reads the claim file?+

Three readers. Your own agents, where you control the context. The buyer's assistant, where you don't, and where what other people wrote about you carries most of the weight. And people, who more and more read what the first two produced. Since November 2026 the second reader also has a profile, a product feed, and a business agent you can claim in ChatGPT.

Why does the claim file sit under the CPO?+

Because the question that keeps it current is asked at merge time, by people looking at the diff, and the agents that load it ship through the product pipeline. A file owned outside that pipeline gets copied into prompts once and drifts. Product marketing writes the file and owns the claim, inside the product org, and the release carries it to every surface that reads it.

Related reading

Deeper essays and other handbook chapters on the same thread.

Part of

Product Leadership