Most investors will look at several hundred pitch decks this year. A growing number of those decks were built by AI. A growing number will also be screened by AI.
So what, exactly, is a pitch deck still for? This month, in collaboration with Commonplace, we take a look at that question.
A pitch deck has always done more than present an opportunity. It requires a founder to articulate what they are building, for whom and why it matters. How big is the market, what is the business model, why this team and why now? Getting those answers onto ten or twelve slides, in a way that an investor would find credible, was hard.
Investors read the deck on two levels. They read for content: does it align with their thesis, do they find the market exciting. But they are also reading to understand the founding team. Papermark's 2026 Fundraising Report, measured across 24,541 decks and 358,672 investor views, shows the team slide holds the most attention of any slide type at 5.7 seconds per view. A founder who could put a clear, well-structured case onto a small number of slides was showing you something about themselves, as well as the opportunity.
The reason this worked is that the two are connected. The signal investors read for in a pitch deck was only meaningful because the deck took real effort to produce. Large language models have made that effort largely optional. At the same time, a growing number of investors are using AI to triage those decks before a human ever reads them. When what goes into a pitch deck and how it is reviewed are both changing, is the investor screening process still working the way it should?
Putting pen to paper is now easier
Forming sentences, arguments and structured presentations can now be done in seconds. Gamma, one of a growing number of AI presentation tools, can produce a complete deck in under sixty seconds. Many founders are using these tools to go beyond arranging slides: generating narrative structure, market sizing, competitor analysis and financial projections. At pre-seed, the deck can end up looking like the most polished thing about the business.
A well produced deck used to be hard for a reason, and that difficulty was what made it a signal. A founder who genuinely understands their market and business model will still produce a strong deck. But now, so could a founder who does not. The visible difference between the two has narrowed.
Many investors have noticed the noise in the signal and are pushing back. Will Richardson, Managing Partner at Giant Leap, wrote that LLMs are like "the greatest cover band in history" (flawless but derivative), arguing investors are craving original thinking over polished AI prose. But asking founders not to use AI assumes the answer is individual restraint. When a tool reduces effort and adoption is widespread, the question is not whether people will use it but how quickly.
There is an irony here. A founder who does choose that friction, who takes the time to structure their own reasoning rather than generating it, may be showing you something genuinely valuable about how they operate. But the problem is detection. Today, an investor might still spot the difference because we are getting attuned to AI slop. As models improve, that will become harder, and you cannot build a screening process around a signal you can no longer reliably identify.
The signalling problem is not only about what a deck implies about the founding team. The numbers have always been the founder's chosen presentation of their own company, and this challenge predates AI. At pre-seed and seed, investors are seeing a ten to 12 slide teaser with high-level, founder-prepared figures, with no detailed model to verify against and no data room to cross-reference. A pitch deck has never been an independent assessment of a business. It is a selling document. This is part of why screening is so quick. Papermark's data across 358,672 investor views puts the average at four minutes, but that average hides a wide variation: 16% of views end within 10 seconds and fewer than half of investors reach the final slide.
When both sides use AI
The other side of the table is changing just as fast. Many investors now use LLMs to extract data from each deck, score it against their criteria and sort it into a shortlist. In some cases a human reviews that shortlist. In others, the model's output is the decision. Affinity's 2026 Predictions Report found that over 80% of data-driven VCs using large language models have incorporated them into sourcing.
So, in a growing number of cases, founders use AI to build the deck and investors use AI to read it. The document becomes a transaction between two models. This is already producing unexpected consequences in adjacent fields. Researchers at Duke University, working with the recruiting platform HireEZ, analysed around 200,000 real resumes and found that roughly 1% contained hidden text designed to manipulate AI screening. Most of this was invisible 'data injection', but a subset included explicit instructions in white-on-white font telling the system to treat the candidate as highly qualified. A pitch deck is not a CV, but it does the same first-screen job for a company.
Prompt injection is not the only issue. The bigger question is what LLMs are screening for and whether that is, in fact, the right thing to screen for. If the task is checking factual fit at a high level, does the company match the fund's thesis, stage and sector, then automation may be an effective first filter. But when LLMs are used to assess the quality of an opportunity, the risk changes. Early stage investing, and much of later stage too, is a bet on the unexpected. LLMs are trained on patterns that already exist, they do not predict what has not happened before. The risk is that the best opportunity in your email or WhatsApp gets sorted into the reject pile by a model that cannot recognise what makes it different. And most investors would probably never know. Very few go back and audit what was rejected.
Is it time to change the way we screen?
A pitch deck has always tried to do multiple jobs: fit check, data presentation, and founder signal. The fit check largely still works. Thesis, stage, sector, geography, cheque size. That filter is binary and quick and automating it can make sense. But it does not need a pitch deck. AI has weakened the deck's ability to do the other two.
At pre-seed and seed the investment is almost entirely in the founder and the idea. If that signal is now less reliable, the question is whether the deck is still the right starting point. Y Combinator does not ask for one. Their application is text-based with a short video, no slides. The interview is ten minutes of direct questions testing whether the founder can think clearly under pressure.
As a company moves toward Series A, real data begins to exist and the separation becomes practical. The analytical work can be done independently, and the founder conversation can focus on the harder questions. Parker Conrad raised a $45M Series A for Rippling in 2019 with no traditional pitch deck: an 11-page investor memo in prose, accompanied by a separate 46-slide metrics deck.
The deck worked because it bundled everything into one format. What comes next could be a separation: structured data that can be analysed independently and a quick human touchpoint that tells you something about the founder that slides do not. The two jobs are pulling apart and the investors who recognise that early and adopt new processes may find the best new opportunities before anyone else. Of course, until investors stop asking for them, founders will keep creating them. The change has to come from both sides of the table.
How Allermuir approaches it
Hebrides, Allermuir's valuation platform, is built for that separation. It handles the analytical job: retention, margins, comparable companies, a valuation range, all tested independently of whatever narrative a founder presents. It draws in external data sources to validate and pressure-test the numbers, acting as a thinking partner for the investor rather than a replacement for their judgement. When the cost of building software is this low, more companies will be built by more people, and the team behind the company becomes an even greater differentiator than it already was.
When the analytical work is already done before the first meeting, the conversation changes. An investor is no longer asking a founder to walk them through the numbers. They are asking them to explain the gaps, defend the assumptions, and show why they are the right people to build this. That is a better conversation. And it is a better use of the thing that no tool can replace: the investor's own judgement.
Commonplace is the private, curated community for LPs, built around the idea that the best conversations in private markets happen off the record. We're grateful to Jocke and the team for the collaboration.