Opening

You briefed a marketer to write copy for your product. They delivered. The copy is accurate, well-structured, tonally right. You put it live. Six months later you run your product through an AI assistant - the same assistant your prospects are now using to research purchases. The AI draws a blank, or worse, returns a competitor. The campaign worked. The interpretation layer missed you entirely. The brief was the problem. Not the copy.

The_Brief_Was_Written_for_One_Reader

The marketing brief has always been a transfer document. Its job is to move context from the person who knows the product to the person who will write about it. Here is what we are building. Here is the audience. Here is the emotional register. Here is what we want people to feel. Here is the objective.

That logic was complete when the output had one reader: a human being who would encounter the copy on a page, in an email, in a social post, and respond to it as a human being. The brief specified what the human reader would experience. The marketer produced it. The brief's work was done.

The assumption that underpinned it - that the output's reader is human - held for decades. It does not hold now. Not because the human reader has gone anywhere, but because there is another reader between the marketer's output and that human. A reader that processes the copy before any person opens the page. A reader that decides what is a provable claim and what is noise. A reader that has no patience for warmth, no capacity to respond to tone, and no interest in what the brand wants to project.

Most marketing briefs have not noticed this reader exists.

The_Interpretation_Layer_Arrived_Before_the_Brief_Updated

AI-mediated research is not a shift to prepare for. It is the current state for a significant share of purchase decisions. Buyers use AI assistants to build their consideration set, filter options, and assess credibility before they have opened a single vendor page. For B2B purchases, Forrester's 2024 research found 89% of B2B buyers use generative AI for self-guided research; 6sense's B2B buyer experience research put roughly 70% of the decision journey complete before the first direct contact.

The implication for a founder who commissions marketing work: the marketer's output does not arrive directly in front of a buyer. It arrives in front of a retrieval system that decides what to surface. That system extracts what it can cite. It matches claims to the terms buyers use when they search. It ignores what it cannot parse as a specific, verifiable statement.

The marketer cannot compensate for this with better copy. Better copy in the old sense - more compelling, more beautifully written, more tonally precise - makes no difference to a system that is not reading for tone. The marketer can only work with what the brief specifies. If the brief does not specify "this claim must be stated as a verifiable mechanism," the marketer produces a claim that is emotionally resonant and extractably invisible. Not a failure of execution. A failure of specification.

This pattern appeared earlier in the product brief - and in what it means to brief agentic systems rather than interface-based ones. That argument is in The Brief Is Changing. What is new is that the same structural shift has arrived at the marketing brief, where you are probably not expecting it yet.

What_the_Second_Reader_Actually_Needs

The interpretation layer is not a reader in any useful sense of that word. It does not respond to voice, warmth, brand personality, or the careful calibration of register that good copywriting produces. It responds to something structurally different: claims it can extract and attribute without guessing, evidence it can cite without introducing its own inference, and phrasing that maps to the terms buyers actually use when they search.

Three things the interpretation layer can work with:

Structured, machine-extractable claims about what the product does. Not "a smarter way to manage your workflow" - what it does, stated as a specific mechanism or outcome. The system cannot extract "smarter." It can extract "reduces the time to generate a weekly report from 45 minutes to 8 minutes, by automating the data pull and templating the structure." One of these survives the extraction test. The other is invisible.

Evidence that can be cited without interpretation. The interpretation layer favours claims it does not have to verify by trusting you. A number, a comparison, a mechanism, a test result - things a system can point to and say "this is the basis." "Best in class" is invisible. "Tested against six competing tools in a blind evaluation by 40 practitioners, rated highest for output accuracy" is citable. The difference is not modesty or precision - it is whether the claim is architected for the second reader.

Specificity that maps to how buyers actually search. Buyers do not search in brand language. They search in problem language. "How do I reduce the time my team spends on weekly reporting?" is a real search. "Smarter workflow management" is not. A brief that does not specify the buyer's actual query language cannot produce copy that surfaces for the query the buyer is running.

An emotional brief produces emotional copy. Emotional copy fails the extraction test. The brief is the upstream cause.

The_Old_Brief_vs._The_New_Brief_-_Side_by_Side

The clearest way to show this is to put two versions of the same brief next to each other. The product is the Labs site - labs.dianaismail.me - which I can use because I have written both versions of its brief, and because the contrast between them is the argument in documentary form.

Here is the version I would have written two years ago:

Working title: Diana Labs. Core message: I am continuously learning. Not in theory. In practice. Audience: Potential employers, collaborators, recruiters, creative technologists, innovation teams, marketing leaders, and agency people who want to understand how I think. Tone: Direct, reflective, practical, a bit funny, comfortable admitting that not everything is fully figured out. Confident but not inflated. Brand personality: Experimental, slightly edgy, curious, intelligent but not academic, human, occasionally weird, practical underneath the chaos. What the site should show: That I am not just talking about AI from a distance. I am actively exploring it. Testing it. Building small things.

That brief is well-written. It transferred context accurately from my intent to anyone who would execute it. It would have produced copy that was tonally right, personality-consistent, and extractably invisible.

The failure mode is named in the brief itself: "I am continuously learning. Not in theory. In practice." That sentence - which was the site's core message - is exactly what the interpretation layer filters past. It is a declaration with no evidence attached. A retrieval system cannot cite it. A buyer vetting me through an AI assistant gets nothing they can hold.

Here is the version I wrote when the second reader became a build requirement:

What this site is: The standing record of the work I am still doing. Labs exists to be the dated, accreting evidence that I am working at the edge of this continuously - not on a CV, but in public, with artefacts. Audience: A person, and the retrieval system deciding whether to cite the person. Claims Layer: A structured registry of extractable claims - specific, true statements I want an engine to be able to extract and attribute. Each claim carries its evidence tier (Tier 1: externally verifiable; Tier 2: first-party with artefact; Tier 3: labelled position). Evidence Tier: Every claim publishes at its true strength. A claim with no artefact is not a fact yet - it is a Tier 3 position, and it publishes labelled as one. Buyer-Query Terms: Nobody types "scope fluency" into Perplexity. Buyers type the problem. The query layer maps the language they actually use to the content that should surface for it.

Five structural differences. Not style differences - architectural ones.

The before-brief names one audience: human readers. The now-brief opens by naming two: "a person, and the retrieval system deciding whether to cite the person." The before-brief's core message is a declaration ("I am continuously learning"). The now-brief opens by diagnosing that declaration as the failure mode ("A page that says 'I'm always learning' is the failure mode - credential-free and an AI engine filters past it the same way a Brand VP does"). The before-brief has positioning statements and tone direction. The now-brief has a Claims Layer - a structured registry of extractable claims I am building the site towards, each with evidence tier and artefact. The before-brief has no concept of evidence gradation. The now-brief makes tiering non-negotiable. And the before-brief has no concept of how buyers search. The now-brief has a full Buyer-Query Terms section built around the recognition that owned vocabulary ("scope fluency") is not search vocabulary.

The difference is not that the second brief is better written. It is that the second brief specifies output for two readers, not one.

The_Brief_Is_Now_Two_Documents_Stacked

A fit-for-purpose marketing brief now has two layers that must coexist in the same document.

The first layer is the one founders know how to write. The human spec: tone, audience, structure, channels, creative direction. The emotional register the brand wants to project. The voice the marketer needs to match. This layer is still required. A marketer who receives only a claims registry and no human spec will produce structured content that is factually extractable and tonally inert. Both layers are necessary. Neither replaces the other.

The second layer is the one most briefs are missing. The machine spec: the provable claims layer. Specific product claims stated as verifiable mechanisms or results. The evidence that makes each claim citable - numbers, comparisons, mechanisms, tests. The query language buyers actually use, mapped to the product's answers. This layer is what the interpretation layer reads. Without it, the marketer cannot produce output that survives the extraction test, regardless of how well they execute the human spec.

The brief-writer's job is now to produce both, explicitly, or to make it part of the deliverable that the marketer produces the machine spec as part of the commission. The marketer may be better positioned to write the buyer-query map than the founder - they often have closer contact with how buyers describe the problem. But someone has to specify it. If the brief does not, it will not happen.

The brief is not longer. It is differently shaped. It answers two questions: what should this copy make a human feel, and what must this copy make a machine able to extract and cite?

The_New_Brief_Skeleton

Not a principle. An artefact. Four fields the brief needs that most briefs do not have. Filled here with Geoverity - a tool I built to run a brand through AI search surfaces and return a structured gap analysis.

PRODUCT CLAIM (provable). What it does, stated as a specific mechanism or result. Not positioning - mechanism. Example: Geoverity (geoverity.io) runs a brand through AI search surfaces (ChatGPT, Perplexity, Google AI Overviews, Meta AI) and returns a structured gap analysis: which claims are being cited, which are being missed, which are being attributed to a competitor.

EVIDENCE TIER. The data, comparison, or test that makes the claim extractable. Assign the correct tier: T1 = externally verifiable (public repo, named third-party study, live URL anyone can run); T2 = first-party with artefact (dated build log, usage data, running system); T3 = labelled position (judgement, hypothesis, opinion without external backing). Example: T2. The audit mechanism is first-party - the running system. No conversion claims; no live-user figures. The claim is what it does, not what it produces.

BUYER QUERY MAP. The terms buyers use when searching for this problem. Not the product's vocabulary - the buyer's language. Example: "does my brand show up in AI search results" / "how do I know if ChatGPT knows my product" / "why isn't my company appearing in AI-generated answers" / "how to audit AI search visibility" / "brand visibility in ChatGPT Perplexity Gemini".

HUMAN REGISTER. Tone, voice, emotional register - for the marketer, not the machine. Example: Direct, builder-facing, technically specific without being academic. The reader is a founder or a marketing lead who already suspects there is a problem and wants to know what to do about it. Respect that. Do not over-explain.

Fill these four fields before handing the brief to a marketer. The marketer still needs the rest of the brief - the full audience profile, the channel specs, the creative direction, the deliverables list. But these four fields are what the brief currently does not have, and their absence is where the interpretation layer loses you.

The brief has two readers now. Write for both.