AI Due Diligence: What ChatGPT and Perplexity Tell an Investor Who Screens You Before an IPO or M&A

Before booking the first call, an investor increasingly opens ChatGPT or Perplexity rather than Google and asks: "What is known about [name]?" The model's synthesized answer — with its errors, its tone, and its choice of sources — becomes your dossier before due diligence even begins. A GEO audit of your digital profile shows you that answer before your counterparty sees it.

"What does ChatGPT say about me" is no longer idle curiosity. For anyone preparing a company for an IPO, selling a stake, or raising a round, it is a question about the dossier the other side will bring to the table. Below: how this new layer of first impressions works, what to check inside the AI's answer, and how it differs from a classic digital-footprint review.

The first screen is now run by a model, not an analyst

A couple of years ago a pre-deal reputation check began in an analyst's hands: they opened a search engine, scrolled the results, and gathered links to articles, court records, and social media. Today the first pass is increasingly done by a generative engine. A fund analyst, a partner ahead of the investment committee, an underwriter's team at the bank — any of them may start not with ten Google tabs, but with a single query to ChatGPT, Perplexity, or Gemini.

The difference is fundamental. A search engine returns a list of links and leaves the conclusion to a human. A model returns the conclusion — one coherent paragraph in which facts, rumors, stale news, and guesswork are already blended into a single "verdict" and delivered in a confident tone. That is the new first impression: not ten sources to weigh, but one synthesized answer that reads like an objective brief, even though it is not.

For a deal, that raises the stakes. The same digital footprint that used to move the M&A price through manual analysis now passes through the model first — and reaches a human already digested, with emphases you never controlled.

Five questions an investor asks the AI about you

The good news: this screen is reproducible. You can sit in the seat of whoever is checking you out and ask the model the exact same questions. Below is a baseline self-check protocol. It needs no special tools — just access to the engines and honesty in how you phrase things: ask the way a skeptical investor would, not the way you wish you'd be described.

1. "Who is [name / company]?" — the baseline portrait: what the model treats as central about you and where it starts the story.
2. "Has [name] been involved in lawsuits, claims, or regulator action?" — a check for negative and legal context.
3. "Is [name] linked to sanctions, PEP status, or investigations?" — the risk layer most sensitive to a deal.
4. "What is the reputation of [company]?" — how the model summarizes reviews, coverage, and mentions of the business.
5. "Should I invest in [project] founded by [name]?" — a direct investment question on which the model issues a value judgment.

Run it across five systems

One engine is one opinion. The full picture comes from running the same questions across several systems at once: ChatGPT search, Perplexity, Gemini and Google AI Overviews, and Copilot (for CIS-based investors, add Alice / YandexGPT). Each relies on its own set of sources and its own synthesis logic, so the answers will differ. Those discrepancies are the most valuable output of the self-check: where the engines contradict each other is exactly the zone an investor will notice and start to probe.

What to check inside the AI's answer: five audit axes

Getting the engines' answers is only half the job. Next you have to break them down along axes, not read them as prose. The same sentence from a model can be harmless on facts yet toxic in tone; technically correct yet dangerously incomplete. A GEO audit decomposes the answer into five independent dimensions — and any one of them can decide how the first conversation goes.

An empty profile is a red flag too

Instinctively, the best audit result seems to be when the model "says nothing bad" about you. In practice, for an investor a void is often worse than measured negativity. An answer like "there is not enough data on [name]" reads as opacity to the person across the deal: a public figure of that scale should leave a trail, and its absence prompts the question "what are they hiding?" So "no data" is not a neutral result but an alarming one, and it deserves as much attention as outright negativity.

How a GEO audit differs from classic due diligence

A GEO audit does not replace a classic digital-footprint review — it looks at a different surface. Classic due diligence works with sources: media, courts, compliance databases, social media. The GEO layer works with what the model synthesized from those sources and delivers as a finished answer. The difference is in the table below.

See also · IPO & Investment What investors check in a classic digital footprint before an IPO The four layers of manual review — media, courts, compliance databases, social media — that underwriters go through before approving a deal. Read the article →

Three ways a finding in the AI's answer hits a deal

An abstract "risk in the AI layer" sounds unconvincing until you see how it plays out. Here are three typical scenarios in which the model's answer does damage before you even learn there's a problem.

A lower valuation. An analyst runs the founder through a model, gets an answer that leans on an old conflict, and bakes that "risk factor" into the valuation model. Nothing has been deleted or fabricated, yet a discount to the price is already there — because the tone of the verdict did its work.

Trust broken on the first call. A fund partner opens the conversation having already "read" you in ChatGPT. If the model's answer diverges from what you say about yourself, a sense of mismatch sets in — and the conversation proceeds from a position of distrust, even though the discrepancy may simply be a model error.

It surfaces in the investment memo. A phrase from the AI's answer migrates almost verbatim into the fund's IC memo — and there it takes on a life of its own, shaping a committee decision you have no direct access to.

When it's not the truth but a mix-up

A separate and insidious case is when the negativity in the answer doesn't concern you at all. The model confuses you with a namesake, attributing someone else's convictions, bankruptcies, or scandals to a person with a similar name. To an investor it is indistinguishable from your own footprint; to you it is the most galling failure of all, because you're "correcting" something you never did. In an audit, this false association must be identified separately from genuine negativity — the routes for handling them differ.

What can be fixed before a deal, and what's already too late

The main constraint of a GEO audit is time. You can only influence the model's answer indirectly, through the primary sources it assembles that answer from. And any change to those sources reaches the models with a lag — weeks to months — as data is re-crawled and folds into the next updates. So what you can actually get done depends directly on how many weeks remain before the deal.

If you have runway, here is what that time can buy:

1. refresh and strengthen authoritative primary sources — publications, profiles, expert material the engines lean on;
2. put the knowledge panel and structured data about the person and the company in order;
3. resolve any namesake confusion, separating your entities in the eyes of models and search engines;
4. fill the "voids" — the facts and achievements whose absence reads as opacity;
5. build up fresh, authoritative mentions so they outweigh stale or accidental content.

What you cannot do a week before the call is rewrite the model's settled verdict "for the deal." If the AI already confidently repeats an error, there simply isn't time for new data to reach it. A separate story is when the AI is already saying something false about you: that isn't deal prep but reactive correction of a hallucination, and it should have started earlier still.

Service Digital Profile Creation Let's talk →

The layer most sensitive to a deal deserves its own note — the sanctions and PEP context. If the model links you to sanctions lists or politically exposed persons, even mistakenly, that can halt a deal faster than any negative review, and such a signal must be dealt with first.

What we don't promise

Let's be clear about the limits. We can improve the input data a model builds its answer from — but we cannot "order" a model to say what you'd like. ChatGPT, Perplexity, and Gemini have no "delete me" button, and any agency promising to "erase you from ChatGPT" by a fixed date is selling an illusion. Models change constantly: an update, fine-tuning, or a shift in sources can rewrite the answer in any direction. So a GEO audit isn't a one-off exercise but ongoing work: it's worth checking the AI layer not only before a specific deal but systematically, because the next investor will put their question to a different version of the model.

Frequently asked questions

What to do next

If a round, a stake sale, or an IPO is ahead, start with a simple step — put the protocol questions above to the models and look at the answer through an investor's eyes. That alone shows whether your AI layer holds errors, gaps, or an awkward tone. Next, decompose the answer along the five axes and decide what can realistically be fixed in the time left before the deal — and what you'll have to walk into the conversation with as is.

Aegis Reputation Lab runs a GEO audit of your profile across every key engine, confidentially — before your counterparty does. Get in touch to discuss your situation and your deal timeline.

See also

IPO & Investment

Digital Footprint Before an IPO: What Investors and Underwriters Check

Read the article → M&A and Deals

Due Diligence When Selling a Business: How the Digital Footprint Affects the M&A Price

Read the article → AI Search

Reputation in AI Answers: What to Do When ChatGPT Gets It Wrong

Read the article → AI Identity Confusion

AI Entity Confusion: When ChatGPT Mixes You Up With a Namesake

Read the article →