July 22, 2026

Can ChatGPT Write a Pitch Strategy? We Ran One Brief to Find Out

We've said AI builds the slides of a pitch but not the strategy, and that the only way to judge a tool is to see its output for yourself. So this time we ran the test: one campaign brief, put through a general-purpose AI and through Synapsis. No arguing — just the results, side by side, and what was different.

One caveat up front: this was a single informal test, not a controlled study — one brief, one run per prompt level. We're sharing what we saw, not a benchmark. Judge the output logic, not the sample size.

How the test was set up

We kept it as fair as we could. Same brief for both — built from the real profile of a food-delivery app brand, with the same five fields entered each time: client, objective, background, target, and core message.

The comparison was a widely used paid chatbot — the kind most teams reach for first. But we split the prompt into two levels. One was the bare minimum you'd type when you're busy: "Using this brief, write the strategy section of an ad pitch, three different directions." The other added format on top: "Write three strategic directions for an ad pitch, each a different approach, around 25 slides each, with a core argument on every slide."

The only difference between the two is length and format. Both asked for three distinct directions. That matters, because in practice people are usually in a hurry when they hand work to AI — you're using it to *save* time, so you don't spend much polishing the prompt. And creative strategy is a domain where you have to already know what's persuasive to instruct it well, which makes a precise prompt hard to write in the first place. So we treated those two levels as a realistic ceiling and floor.

The bare prompt: a well-written shell

Even without any format spec, the general-purpose AI produced a plausible-looking pitch. There was a market read, a target insight, an idea, an execution plan, a close. The sentences were clean. At a glance, nothing wrong.

Then we went slide by slide, and the problems surfaced.

First, no evidence. "Delivery fees keep climbing, so people order less," "diners are tired of comparing apps for the cheapest delivery" — every line is plausible, and not one has a source behind it. Nothing supporting any claim. It sounds right; there's no way to check whether it *is* right.

Second, it invented KPIs. Numbers like "awareness up X%, conversion up Y%" appeared, with no basis for where they came from. KPIs don't even belong in the strategy section of a pitch — the model saw the word "pitch" and bolted on figures no one asked for.

The formatted prompt: it hit the format, and then

The second run was the more interesting one. Spell out length and structure — "around 25 slides each, a core argument on every slide" — and you'd expect a better result.

The AI nailed the format. Three directions, 25 slides each, one line of message per slide. It looked exactly as instructed.

But from about the tenth slide of each direction, the content started to come apart. The first nine slides passed for strategy; after that it filled up with execution ideas — TV spots, out-of-home, influencer plays. Unable to carry one strategic direction across 25 slides of logical structure, it padded the length with the easy stuff: execution.

And it wasn't just one direction — all three broke at the same point, the same way, which is what made it worth writing up. We'd asked for *strategic directions*; what came back was "a one-line concept with a list of execution ideas stapled to it."

What was different

Run the same brief through Synapsis and the output differed in three specific places.

Claims carried evidence. Each slide's argument was tied to checkable data — supported statements, not invented ones.

It stayed in the strategy lane. It dealt only with strategic direction and never drifted into execution plans or KPIs. It held the line on where a strategic direction ends.

The logic ran end to end. One chain, from situation to problem, through evidence to insight, into strategy. Not a list of execution ideas, but a case for *why this direction*.

Why the difference happens

The reason is the limit of prompting. Writing the strategy section of a pitch isn't like a coding task. With code you can specify exactly what you want; with "what makes a persuasive strategy" you can't — not clearly, not in a few lines. To do that you'd have to already be someone who knows the judgment. What reads true and what's a leap — that call doesn't transfer through a prompt.

No matter how precise your instructions to a general-purpose AI, there's judgment the instructions can't carry. So it mimics the format, invents the evidence, fails to hold the boundary of strategy, and — unable to develop a direction all the way through — collapses into an execution list.

Synapsis builds that judgment — the kind you can't hand off in a prompt — into the engine. Which means the user doesn't need to know it: enter the four brief fields and that's it. No prompt-crafting required, because the judgment already lives inside the tool.

A pitch already costs a team weeks of work, and you don't win them all. The last thing you want from AI is a draft that looks done and has to be rebuilt from scratch. Slide design is already fast, thanks to the automation tools. It's the content that goes inside — the strategy — that's due to get faster next.

Why AI builds the deck but not the strategy: AI writes the deck. Who writes the creative strategy?
Why a polished AI draft still fails in a real pitch: Why AI strategy looks great on paper but fails in the pitch
How the engine turns a brief into three directions: Can AI write a creative strategy for a pitch?

Frequently asked questions

Can a general-purpose AI like ChatGPT write the strategy for a pitch?

It can produce a formatted document. But in our test, the general-purpose AI stated claims with no evidence, couldn't develop a strategic direction fully and slid into execution plans and invented KPIs, and padded length with lists of execution ideas. It looks like a strategy pitch; to use it in a real one, a person rebuilds it from scratch.

If I write a better prompt, will a general-purpose AI give the same result?

In our test, even a prompt that specified length and format only got the format right — the content still wasn't usable. The judgment behind "what makes a persuasive pitch strategy" is hard to move into a prompt. To express that judgment as instructions, you'd have to already hold it yourself.

How is Synapsis different from a general-purpose AI?

Synapsis is built to produce only the strategy section of a pitch. Each claim is tied to checkable evidence, it holds the line at strategic direction, and it builds a logical chain from situation through insight to direction. You enter four brief fields and nothing more.

What does a general-purpose AI get wrong in a pitch strategy?

Evidence and logic. It generates smooth sentences, but there's no source backing the claims and the case for *why this strategy* is thin. In testing, even when we asked for strategic direction, it repeatedly substituted a list of execution ideas instead.