There’s a chart in one of our brand-intelligence reports whose axis label reads “Queries won against you (measured).”
It ranks the rivals AI engines actually cite on the queries where your brand loses, ordered by how many each one wins. No strategist assembles it. No model dreams it up either. The code that builds it carries a comment saying the data is computed from measurements, “never LLM-emitted.” And when the landscape hasn’t been measured yet, the chart refuses to render; the prose names the gap instead. The writing model’s instructions are blunt: “never invent a competitor, a citation, or a figure.”
That chart is a small thing. The reason it can exist is the big thing, and it’s the reason I’d push any brand operator to rethink how their competitive knowledge is stored, whether or not they ever use software like ours.
A deck has one consumer
Think about the last competitive analysis your team produced. Odds are it was a deck. Someone smart spent two weeks on it. It named the threats, mapped the positioning, maybe even landed a genuinely sharp insight about why a rival keeps winning a channel you thought you owned. Then the meeting ended.
However sharp the analysis, a deck has one consumer — a human, once. Nothing else in the building can read it. Your media buying can’t, and neither can your content pipeline. The AI tools you’re starting to use can’t read it either, which means every one of them operates in a world where your competition doesn’t exist, or worse, a world where the model fills in your competitive landscape from its training data, the averaged, outdated, sometimes invented version of your category.
That’s the quiet failure mode of competitive strategy in most consumer brands. It isn’t that the thinking is bad. It’s that the thinking is stored in a format only humans can consume, at exactly the moment when more and more of your brand’s output is being produced by things that aren’t human.
However sharp the analysis, a deck has one consumer — a human, once.
What a feed looks like instead
Competitive analysis at our company is a set of structured fields on the brand’s canonical record — the named enemy, the positioning wedge, the competitor frame — each one a value a chart can read, not a paragraph about competitors. The difference sounds bureaucratic until you watch what it enables.
The code that builds that chart declares which of those fields it reads, the way a recipe lists its ingredients up front. There’s a literal list, written into that report: five fields, and only five. The types of competitor the brand faces. The competitors themselves. The enemy. The wedge. The differentiation matrix, which is the grid of what each rival claims against what you claim. The comment beside the list calls it “the brand’s own competitive frame”: the brand’s opinion of its rivals, kept deliberately separate from what the engines measured about them. The report is told which is which.
That declaration matters more than it looks like it does. It means you can ask, of any surface the brand produces, exactly which pieces of competitive truth it drew on. It also means that when a field changes — the wedge sharpens, a new rival enters the frame — every reader of that field can pick up the change without anyone re-briefing them. Nobody re-briefs anybody.
And the consumption compounds. The same fields feed the prompts that classify how each rival wins its citations. The taxonomy has four kinds of win: retailer breadth (cited because it’s stocked everywhere), clinical or ingredient authority (cited for studies and proof), community (cited through reviews and social), and editorial voice (cited by press and expert roundups). Each classification has to be tied to the specific measured gap or competitor domain it explains, so a rival is never labeled “clinical authority” because that sounds right.
The brief that comes out the other end terminates in actions with owners: out-position the top rival on the answers it’s winning from you (brand lead), earn citations on the single source AI leans on most where you’re absent (content lead), correct the one thing the engines get wrong about you, and re-measure monthly (founder), because the position is only as current as the last audit. Those moves are seeded into the products that execute them. One record, many readers: the chart reads it, the classifier reads it, the action brief reads it, and any future surface we build will read it too, for free.
A deck can’t compound. Each new use of a deck’s insight requires a human to remember the deck exists, find it, reread it, and re-type its conclusion into whatever tool needs it. Every hop loses fidelity. A field has no hops.
| A deck | A record of fields | |
|---|---|---|
| Who can read it | A human, once. Nothing else in the building can read it. | The chart reads it, the classifier reads it, the action brief reads it, and any future surface we build will read it too, for free. |
| When something changes | Each new use of a deck’s insight requires a human to remember the deck exists, find it, reread it, and re-type its conclusion into whatever tool needs it. | Every reader of that field can pick up the change without anyone re-briefing them. Nobody re-briefs anybody. |
| What each hop costs | Every hop loses fidelity. | A field has no hops. |
Machine-made, and honestly absent
The part of this system I’d defend hardest is what happens when the data isn’t there.
When the competitive landscape hasn’t been measured for a brand, the ranked chart doesn’t render a plausible-looking placeholder. It returns nothing, and the surrounding prose names the gap instead. There are exactly two conditions under which it goes quiet: the landscape hasn’t been measured, or it has been measured and no rival won a single losing query. Both are real answers.
The absence travels all the way through. Where a number is missing from the signal handed to the writing model, the line literally reads “not yet measured — labeled gap,” and the model is told to print that line as written rather than estimate around it.
The action plan has its own floor. It needs three grounded moves to exist; if the signal can’t support three, the plan is omitted entirely. The comment in the code says why: there’s nothing honest to prioritise from.
If you’ve spent any time with AI marketing tools, you know why that rule has to be written down. Models are eager. Ask one to describe your competitive landscape and it will describe a competitive landscape, fluently, whether or not it knows anything true about yours. An invented rival in a strategy document is a false input to every decision downstream of it.
So the rule in our reports is structural: measured data renders as measured data, gaps render as named gaps, and the model is never the source of a competitor, a citation, or a number. The chart earns its axis label. “Measured” means measured.
An invented rival in a strategy document is a false input to every decision downstream of it.
Honest, but never neutral
We keep one standing rule on competitive surfaces: honest, but never neutral. In the instructions we give the writing model, that’s two lines a few lines apart — “never assign a win-reason the measured signal doesn’t support,” and “Close with conviction.”
Those two lines are a compressed philosophy of competitive strategy, and they’re worth stealing even if you never automate anything.
You don’t get to explain a loss with a story the data doesn’t back. If a rival is beating you and the signal doesn’t say why, the honest output is “we don’t know why yet,” not a confident narrative.
A competitive analysis that ends in a balanced shrug is stenography. Somebody has to say what to do.
The discipline is doing both at once — conviction downstream of evidence, never instead of it.
There’s a third instruction in the same report I’d add to the pair. The writing model is told not to re-argue the brand’s overall position; a sibling report already owns that, so this one has to earn its place by spending every word on the competitive mechanics: who, on what query, why they win it, and the move that closes the gap. A reader should finish with a named target list, not a second copy of the positioning thesis. Most decks get that backwards.
Most competitive decks fail one side or the other. The rigorous ones end neutral. The confident ones invent their support. Encoding both instructions into the machinery was our way of refusing the trade.
Sensitive by architecture
One more design choice. The public version of the brand record (the version outside tools can request, including the AI agents that read it) is built from an explicit list of what’s allowed out, never a list of what’s blocked. Adding a field to that list expands exposure; everything left off it is, in the words of the comment beside the list, “gated by omission.” The competitor list, the competitor types, the differentiation matrix, and the windows where a rival is vulnerable are all left off. They aren’t hidden behind a lock; they simply aren’t there.
The line is drawn in a specific place, and I think it’s the right one. The brand’s own stance, its wedge and the enemy it stands against, is on the public list, because that is what a brand wants the world to repeat. What the brand knows about its rivals, and where it thinks they’re weak, is not. Your position is a message. Your read on the competition is intelligence, treated as real, sensitive data because it is.
Copy one mechanical detail into any system that holds competitive data. The public version is assembled by picking the allowed fields one at a time, never by taking the whole record and deleting the sensitive parts. When someone adds a new competitive field later, the delete approach leaks it by default; the pick approach keeps it private until someone deliberately lets it out.
A deck never forces you to think about this. A deck’s access control is whoever got the email. Once competitive intelligence becomes fields that many systems can read, you have to decide, explicitly, which systems may read them, and for the outward-facing ones the safest answer is the shortest possible list, built in from the start.
Where a brand operator starts
You don’t need our stack to act on this. You need a record.
Open a document (a spreadsheet is fine) and answer, as fields, not prose: who is the enemy (the thing you’re against, which may not be a company), what is the wedge (the one claim you win on that a specific rival can’t match), who is in the competitor frame and what does each of them actually win on. Keep it current the way you’d keep inventory current.
The concrete version is one sheet with two tabs. The first has one row per rival and three columns: the name, the kind of win it gets (retail breadth, clinical authority, community, editorial), and the specific place you’ve seen it win: a retailer listing, a study, a subreddit, a roundup.
If you can’t fill the third column, leave it blank; a blank is a labeled gap and a guess is a fabricated one.
The second tab is the audit: one row per place your brand produces output (agency briefs, the AI writing tool your team uses, ad platforms, your own people) and one column, can this surface see the first tab, yes or no.
The count of no’s is the size of your problem. Every one of them is producing brand output inside an imagined competitive landscape.
A deck tells your team who beat you last quarter. A feed tells every machine you own who’s beating you right now.
Competitive intelligence, kept as fields
The report described here is one we generate inside Jinn’s brand-intelligence set.
See how Chart worksIsn’t a good briefing doc the same thing?
A briefing doc is a deck with fewer pictures. It still needs a human to carry its contents into every tool. The test is whether a machine can read the enemy field without a person in between.
Won’t structured fields flatten the nuance a strategist brings?
The nuance goes in the fields. “Wins on clinical authority via an ingredient study that roundup sites cite” is more nuanced than most decks manage, and a machine can act on it. What fields kill is the paragraph that sounds insightful and commits to nothing.
A deck tells your team who beat you last quarter. A feed tells every machine you own who’s beating you right now.
The report described here is one we generate inside Jinn’s brand-intelligence set; it isn’t something customers can order on its own today.