Check a pitch deck before the partner meeting

It's Sunday evening and the partner meeting is at nine. The deck puts the market in the tens of billions, cites "industry reports" in six-point type, and names a customer you'd love to believe. The memo you're writing will repeat those numbers under your name. This page shows how to check the claims in a pitch deck or memo against filings and published reports, what CiteJury gives you back, and where public sources run out and the data room has to take over.

Why deck claims deserve a second look

Most deck errors aren't fraud. They're stale figures, a projection quoted as an actual, the right number from the wrong year, or a research firm's name attached to a figure that really came from a vendor's blog. They survive because they look sourced, and nobody on the deal team has time to open every footnote.

The rarer cases are the expensive ones. In June 2024 the SEC charged the founder of Joonko, an AI recruiting startup, with defrauding investors of at least $21 million. The SEC alleged that the founder told investors the company had more than 100 customers, including Fortune 500 companies, and more than $1 million in revenue, then produced falsified bank statements and forged contracts when one investor grew suspicious. In April 2025 federal prosecutors charged the founder of the shopping app Nate, which had raised over $40 million on the promise of AI that completed purchases automatically. According to the indictment, as reported by CBS News, hundreds of workers in call centers in the Philippines and Romania did that work by hand.

No public-source check catches a forged bank statement. But the habit that catches ordinary errors, opening the footnote and matching the number to its source, is the same habit that makes the unusual ones stand out.

What you upload and what comes back

Upload the deck as a PowerPoint file or PDF, or the memo as Word or PDF, up to 20 MB. You can also paste text. Reading the document costs about 1 credit. CiteJury finds every checkable claim in it, from market sizes and growth rates to named customers, competitor facts and regulatory milestones. You pick the ones that matter, up to 25, and can reword any of them before they run.

Each claim then gets its own check. CiteJury searches several independent search indexes, favoring primary sources such as filings, regulators, government statistics and official pages, and it reads PDFs, including long reports and SEC filings. It looks for evidence against the claim as well as for it. The best passages are sealed into one evidence pack. Claude, ChatGPT and Grok each read it and answer separately, every quote they cite is checked word for word against its source, and a final review writes the verdict.

You get one report for the whole document. The Claims tab lists each verdict with a headline and a suggested fix you can copy. The Document tab shows the original text with every claim highlighted in its verdict color. Download the report as Excel for the deal file.

A Correct verdict still tells you something. This check confirmed two figures from sell-side research, and the safe wording spells out how to attribute them.

All three AIs agreed that both numbers match Goldman's figures as reported by Financial Express. But only that source carried them, so the result tells you to attribute the 22% upgrade and the 2% cut to Goldman Sachs research, as reported by Financial Express. In a memo, that footnote is the difference between "Goldman says" and "a newspaper says Goldman says". The check took 93 seconds.

Right number, wrong year

An error that often slips through a quick read is a real figure attached to the wrong period. We checked a claim of exactly that kind: "India's global capability centres generated about $64.6 billion in revenue in FY2026."

CiteJury searched 79 results. The sources it found included nasscom.in and zinnov.com, the industry body and research firm behind the figures, along with coverage in The Economic Times, ThePrint and YourStory. A post on X and a few pages that couldn't be opened were set aside, and 12 sources were locked into the pack. Claude, ChatGPT and Grok all said Incorrect.

The number itself was real. $64.6 billion is the FY2024 figure, and the estimate for FY2026 is about $98.4 billion. The suggested fix: "India's GCCs generated $64.6B in FY2024, rising to an estimated $98.4B in FY2026 (Nasscom-Zinnov)." The check took under two minutes.

Notice which way the error cut. Quoting the older figure as current understates the market by about a third. Deck errors don't only inflate. They can also hide growth that would have strengthened your own memo. Either way, you want the right year before the number reaches an investment committee paper.

Checking a deck before the meeting

1. Upload the file. A .pptx works, and so does a PDF export, as long as it has a text layer.
2. Choose the claims the thesis rests on. Market size and growth, named customers, competitor funding, regulatory approvals and any "first" or "only". You can pick up to 25.
3. Make vague claims checkable. "A huge and growing market" can't be checked. "The European market for X was worth $N in 2025, per the report cited on slide 4" can. Pull the year and geography from the footnote.
4. Pick a tier. Standard, with all three AIs and a final review, suits most claims at about 8 credits each. Deep searches wider and suits contested market sizes. Quick uses one AI and suits a first pass.
5. Start with the bad news. Read Incorrect first, then Sources disagree, then Can't confirm, and open the sources behind each before you rely on the verdict.
6. Ask follow-ups. "What does the same report give for FY2025?" is answered only from the check's own sources, for about 1 to 2 credits.
7. File the Excel report with the memo, or share a link with your partner. Anyone with the link sees the check and its sources, not your follow-ups.

What to check first

• The market size footnote. Top-down market sizes often cite a research firm's press release rather than the report. Check the year, the geography and the definition, since spending, revenue and "addressable" are three different numbers. Our page on tracing a statistic to its source goes further.
• Named customers and partners. A logo on a slide is a claim. If the only source naming the relationship is the startup's own website, you'll see that in the source list, and it becomes a question for reference calls.
• "No direct competitors." Test it by building a list. A company search has three AIs propose and cross-check companies, each backed by a source. A Standard search costs about 48 credits and takes about 10 minutes.
• Competitor facts. Funding rounds, headcounts and launches are usually public somewhere, and they date quickly.
• Public comparables. Revenue and risk factors for listed peers come from their filings, which CiteJury reads.
• Regulatory status. "FDA-cleared" or "licensed in every state we operate in" is a claim the regulator's own pages can often settle.

For a whole memo or data-room document, the same flow applies; see fact-checking a document.

What it won't do

CiteJury checks claims against public sources. That covers a lot of a deck, but not the numbers that matter most at seed and Series A.

• Private metrics are out of reach. Annual recurring revenue, retention, pipeline and burn live in the data room, not on the web. Those claims usually come back Can't confirm, which is accurate: nothing public confirms them.
• Paywalled research can't be read. If a market figure lives in a subscription report, CiteJury can only use what's public, and it lists the pages it couldn't open.
• Charts aren't read. It checks text, not images or charts, so a number that only appears inside a chart image won't be found. Scanned PDFs without a text layer can't be read either.
• It isn't financial advice. It tells you whether statements hold up against sources, not whether the company is a good investment.
• Verdicts can be wrong. A verdict is only as good as the sources found, which is why every sentence links to one.

On confidentiality, only upload what your NDA lets you share. To run a check, the text goes to the AIs you've turned on in Settings, and search queries based on it go to search services. CiteJury keeps the text of a document, not the file, and doesn't train AI models on your content.

None of this replaces reference calls or a look at the bank statements. It clears the public claims quickly, so the hours before the meeting go on the questions only the founders can answer.

Questions people ask