Pitching when every deck is AI-generated

Every founder now has access to the same tools. The deck that took a week in 2021 takes twenty minutes in 2026, and it looks better than the one that took a week.

That should be good news. Mostly it is a problem — because when everyone's deck is competent, competence stops being a signal.

The baseline moved, and nobody told you

Two things happened at once.

Founders started generating decks. AI-generated decks are now the baseline rather than the differentiator, and a generic one impresses nobody, because investors see dozens every week. The polish that used to separate you now blends you in.

Investors started reading with AI too. A 2026 survey of 300 private capital dealmakers found 85% use AI tools in their daily workflow, including first-pass screening. So an AI writes it, an AI screens it, and the human enters late — reading whatever survived.

Optimising for the machine is a trap. The machine passes you along. The human decides.

What replaced polish

The clearest predictor of a deck that scores well is not design. It is the percentage of claims backed by specific, verifiable data rather than narrative.

Which is a precise way of saying: numbers a stranger could check.

• Not "significant early traction" → "412 wallets connected, 88 came back in week two"
• Not "large addressable market" → "1,400 exchanges in Argentina, we have signed 3"
• Not "experienced team" → "we ran payments infra at Ripio for four years"

An LLM will happily write the first version of each. It cannot write the second, because it does not know your numbers. That asymmetry is the whole opportunity. Every sentence only you could have written is a sentence that proves a human with real information was here.

Say what you do not know

The counterintuitive finding: investors in 2026 explicitly look for intellectual honesty about what you have not figured out. Honesty reads as a credibility signal, not a weakness.

This makes sense once you see it from their side. Every deck claims certainty. Certainty is free — a model generates it on request. Calibrated uncertainty is expensive, because it requires actually understanding your business.

Try a slide most founders never include:

What would have to be true We are betting Argentine merchants will accept 1.5% to settle in dollars same-day. We have 12 saying yes in interviews and 0 paying. That is the risk. The next 90 days are about turning 12 into 3 paying.

Nobody writes that. That is exactly why it lands.

Where AI genuinely helps

Not writing the pitch. Pressure-testing it.

• Adversarial review. "You are a sceptical seed investor. Give me the ten hardest questions about this, then the three reasons you'd pass." Then go answer them.
• Legibility check. Paste your one-liner and ask what the company does. If the answer is wrong or vague, the sentence is broken.
• Compression. Ask it to cut your paragraph in half without losing meaning. Take the cut, rewrite in your own voice.
• Research. Market sizing, comparables, regulatory context. Verify everything — this is where models invent numbers most confidently.

The rule of thumb: use it to attack your thinking, never to produce your voice.

What the memorable pitches have in common

Across the pitches people still cite, the pattern is consistent:

They start with a specific person, not a market. A vivid story about one customer's pain sets urgency in a way "the $40B market" never does. Hearts, then minds, then wallets — emotional engagement first, evidence second, the ask last.

They earn the market slide. Investors discount generic growth stories fast. What travels now is proof that this team and this timing line up, and evidence of live demand rather than projection.

They are short. Ten to twelve slides, one idea each. Confidence looks like leaving things out.

They sound like a person. The founder's actual obsession comes through. This is the part that cannot be generated, and increasingly it is the only part that differentiates.

Selling a solution nobody asked for

Most Solana products have this problem: the user did not wake up wanting a blockchain.

Do not sell the mechanism. Sell the outcome, then let the mechanism explain why the outcome is possible.

• Mechanism: "non-custodial, permissionless settlement layer"
• Outcome: "your money arrives in four seconds and nobody can freeze it"

Then, and only then: "That works because it settles on Solana — which is also why it costs a fraction of a cent."

The chain becomes the reason to believe rather than the thing being sold. Judges and investors who have heard four hundred infrastructure pitches notice immediately.

A practical sequence

1. Write the one sentence. Test it on somebody outside crypto.
2. Write the three numbers you would stake your reputation on.
3. Write what would have to be true, and what you do not know.
4. Now open the deck tool.
5. Have a model attack it. Fix what it breaks.
6. Read it aloud. Cut every sentence that sounds like a press release.
7. Give it to a member who will be rude about it.

Step seven is the one people skip, and it is the only one that reliably works.

Sources and further reading

• Why you shouldn't use AI for your pitch deck — where the baseline and screening findings come from
• YC Startup Library — free, and still the best thing written on this
• Sequoia — Writing a Business Plan
• HubSpot — startup pitch storytelling
• Our own How to pitch, when nobody knows who you are

---

Bring it to a cowork and let the room take it apart. Apply to be a member.