Do Investors Use ChatGPT to Research Startups?
Some investors almost certainly do, but we found no reliable survey of how many. The evidence does show that AI use is common among educated adults and B2B buyers, that AI answers about companies can be wrong or dated, and that official sites supply most citations for company facts. So publish clear facts, check what AI says, and ask a securities lawyer before posting about your raise. 56
- No reliable public survey says how many investors use ChatGPT to research startups, so treat any such number with suspicion.
- Indirect only: 52% of US postgraduates had used ChatGPT by early 2025; 6sense says 94% of B2B buyers use LLMs.
- Official sites supplied 63.4% of ChatGPT citations for company-fact questions in one Ahrefs test, so your own facts page matters.
- Under Rule 506(b) the SEC says there is no general solicitation, so get legal advice before posting about your round.
- AI answers can be stale: only 14.4% of responses from one free, logged-out ChatGPT model contained web-search queries in Ahrefs' tracking; paid and search-enabled modes may differ.
Do investors use ChatGPT to research startups?
Some almost certainly do, but we found no reliable public survey that measures it. Due diligence means the checks an investor runs on a company before investing. Treat any percentage you see for “investors who use ChatGPT for diligence” with suspicion unless it names the sample and the question. What follows is indirect evidence, labelled as such.
Adoption is wide among educated adults. Pew found that 34% of US adults had used ChatGPT by early 2025, including 52% of those with a postgraduate degree and 51% of those with only a bachelor's degree. 6 Among adults aged 30 to 49, 41% had used it. 6
Professional buyers also use AI. 6sense reports that 94% of the B2B buyers it surveyed use large language models (LLMs, the technology behind ChatGPT) in their buying process, and the most common use is comparing vendors. 7 An investor is not a software buyer, so this is an analogy, not a measurement of investors.
What would an investor ask an AI about your company?
We expect the questions to be plain and factual first, then comparative. The prompts below are our modelled guess at what an investor might type. We did not observe any investor typing them.
The split matters. For fact questions, official websites made up 63.4% of ChatGPT citations in Ahrefs' 20,000-prompt US test. For comparison questions the share was 46.4%. 5 The first four prompts are fact-style, so your own pages can answer them. The last two lean on other people's pages.
- What does Example Co do, and who are its customers?
- Who founded Example Co, and what have the founders built before?
- How much has Example Co raised, and from whom?
- Who are Example Co's main competitors?
- Is Example Co legitimate? Are there any red flags or lawsuits?
- Example Co vs Rival Co: which is better positioned?
How accurate are AI answers about private companies?
We found no study of accuracy for private-company facts, so treat any answer as unverified. Three nearby findings argue for caution.
- Many answers skip the web. Ahrefs found only 14.4% of 66.6 million responses from one free, logged-out ChatGPT model contained web-search queries in August and September 2026; paid and search-enabled modes may differ. 5
- Tools can be confidently wrong. The Tow Center asked eight AI search tools to identify news articles from excerpts and found they gave incorrect answers to more than 60% of 1,600 queries, with little hedging. 8 That task was news citation, not startup facts.
- Buyers know the risk. Gartner found 51% of B2B buyers say they are more likely to meet misleading information from GenAI, against 49% who say the same of sales reps. 9
Why do funding facts go stale in AI answers?
Funding facts change quickly, and any answer built from older data can miss your latest round, a new hire or a pivot. Only a dated, official page can tell an assistant what is current. If an assistant searches the web, it can find that page. If it answers from memory, it cannot.
This is our reasoning from the 14.4% search rate above, not a measured error rate. 5 It is the main reason to check, repeat and date what you publish.
What should a startup publish before it raises?
Publish a small set of dated pages that state facts and nothing else. Google says Organization markup has no required properties and can sit on the home page or on one page that describes the organization, with fields such as name, legal name, address, logo and sameAs links. 10 Markup is optional, but a clear page with a visible date is what people and assistants can both quote.
If AI knows little about you, your own pages carry more weight. In an Ahrefs test, a little-known brand drew 82% of its new AI mentions from answers that cited the authors' own pages, while an established brand drew only 6%. 11 That is one small test by a vendor, but it points the same way as the 63.4% figure above.
- A company facts page: what you do, who it is for, founding year, location, legal name and team size range.
- A team page with real names, roles and links to each person's profile.
- A press page that links to independent coverage, not only your own releases.
- A product page in plain language with a “last updated” date.
- Matching one-line descriptions on LinkedIn and the databases investors use, so the same facts appear everywhere.
What rules limit what you can say publicly while raising?
In the United States, the rules depend on how you raise. Under Rule 506(b), the SEC says there is no general solicitation or advertising to market the securities. 1 Under Rule 506(c), general solicitation is allowed, but all purchasers must be accredited investors and the company must take reasonable steps to verify that status. 2
The rule text says general solicitation includes any advertisement, article, notice or other communication published in a newspaper, magazine or similar media, and any seminar whose attendees were invited that way. 3 Whether a given web page, post or interview crosses that line depends on the facts, so ask a securities lawyer before you publish. The SEC also says exempt offerings remain subject to the antifraud provisions of the federal securities laws. 4
Other countries have their own rules, and we have not covered them. If you raise outside the US, get advice for each place where you speak to investors. A practical default is to keep public pages factual and evergreen, and to leave offer terms for private conversations unless your lawyer says otherwise. This is general information, not legal advice.
| Route | Public promotion of the round | Who can invest | Key condition |
|---|---|---|---|
| Rule 506(b) | Not allowed: no general solicitation | Unlimited accredited investors, and up to 35 non-accredited in any 90-day period | Non-accredited investors need enough financial and business knowledge |
| Rule 506(c) | Allowed | Accredited investors only | The company takes reasonable steps to verify accredited status |
How do you check what AI says about you before a pitch?
Run a fixed set of ten prompts in four assistants and save the answers. Use the modelled prompts above plus two about your market. Mark each statement right, stale or wrong, and note the page each answer cites. Repeat before every round of outreach.
Fix your own pages first, then profiles you control, then ask third parties to correct theirs. Our guides to auditing your brand in AI search and fixing wrong information AI says about your brand walk through the steps. The Query Fan-Out Explorer shows what an assistant searches for.
Should you put traction numbers on your website?
Only numbers you can support, and only after legal review during a raise. Public numbers can be quoted by an assistant long after they stop being true, so date them and update them. The SEC says exempt offerings stay subject to antifraud provisions, so inflated claims can be a legal risk and not only a style problem. 4
What is still unknown?
We found no survey of investor behaviour with AI, no accuracy study for startup facts, and no evidence that publishing more pages raises your chance of funding. Chalk Labs has not published results in this field. It has run Google-focused search work for a local shop and on a co-founder's personal site; neither involved fundraising. We do not guarantee any effect on fundraising.
For the wider context, see AI SEO for startups and AI SEO for AI startups. For a founder's own profile, see how founders show up in ChatGPT and Google AI.
Frequently asked questions
Do VCs use ChatGPT for due diligence?
Some likely do, but we found no reliable survey. For professional buyers in general, 6sense reports 94% use LLMs in the buying process, which suggests AI is part of many research routines. 7 Do not rely on a number. Check what AI says about your company and fix errors before you pitch.
Can ChatGPT get my funding details wrong?
Yes, it can. Ahrefs found only 14.4% of responses from one free, logged-out ChatGPT model contained web-search queries; paid and search-enabled modes may differ. 5 The Tow Center found AI search tools gave incorrect answers to more than 60% of queries in a news-citation task. 8 Keep a dated, official facts page and re-test regularly.
Is it legal to post about my raise on LinkedIn?
How do I correct wrong information about my startup in AI answers?
Correct your own pages and profiles first, since official sites supply most citations for company facts. 5 Then ask any third-party site that is wrong to fix it, and re-run your prompts. Our guide on fixing wrong information AI says about your brand lists the steps.
What should an investor be able to verify on my website?
Your legal company name, location, founding year, a plain description of the product, team names with links to their profiles, and links to independent coverage. Add a visible date and update it. Leave out numbers you cannot support, and have a lawyer review pages that touch on your round.
Sources
- Private Placements - Rule 506(b) · U.S. Securities and Exchange CommissionOpened October 20, 2026. Supports: Page last reviewed 21 September 2026: Rule 506(b) has no general solicitation or advertising to market the securities; unlimited accredited investors and up to 35 non-accredited investors in any 90-day period, who must have sufficient financial and business knowledge.
- General solicitation: Rule 506(c) · U.S. Securities and Exchange CommissionOpened October 20, 2026. Supports: Page last reviewed 17 March 2026: Rule 506(c) lets issuers broadly solicit and generally advertise an offering if all purchasers are accredited investors and the issuer takes reasonable steps to verify accredited status.
- 17 CFR 230.502 General conditions to be met · Legal Information Institute, Cornell Law School (reproducing the eCFR)Opened October 20, 2026. Supports: Paragraph (c): no general solicitation or general advertising, including any advertisement, article, notice or other communication published in a newspaper, magazine or similar media, and seminars whose attendees were invited by general solicitation.
- Exempt Offerings FAQ · U.S. Securities and Exchange CommissionOpened October 20, 2026. Supports: Page updated 17 March 2026: exempt transactions are subject to the antifraud provisions of the federal securities laws.
- How ChatGPT Does Query Fan-Out (15 Million Queries Studied) · AhrefsOpened October 20, 2026. Supports: Published 12 October 2026: 14.4% of 66.6 million free, logged-out ChatGPT responses triggered a web search (August to September 2026); in a 20,000-prompt US test, official sites were 63.4% of citations for fact questions (pricing, features, company details) and 46.4% for comparison questions; snapshot of one model.
- 34% of U.S. adults have used ChatGPT, about double the share in 2023 · Pew Research CenterOpened October 20, 2026. Supports: Survey of 5,123 US adults, 24 February to 2 March 2025: 34% had used ChatGPT; 52% with a postgraduate degree, 51% with a bachelor's degree only; 41% of ages 30 to 49.
- 2025 B2B Buyer Experience Report · 6senseOpened October 20, 2026. Supports: 94% of surveyed B2B buyers use LLMs in the buying process; the most common use is comparing vendor offerings; vendor-run and self-reported; buyers are software and services purchasers, not investors.
- We Compared Eight AI Search Engines. They're All Bad at Citing News. · Columbia Journalism Review, Tow Center for Digital JournalismOpened October 20, 2026. Supports: Published 6 March 2025: eight AI search tools, 1,600 queries identifying news articles from excerpts; collectively incorrect on more than 60%; rarely used qualifying phrases; news sample, not startup facts.
- Gartner Survey Finds 69% of B2B Buyers Turn to Sales Reps to Validate AI-Generated Insights · GartnerOpened October 20, 2026. Supports: Survey of 645 B2B buyers released 20 May 2026: 51% say they are more likely to encounter misleading information from GenAI; 49% say the same of a sales rep.
- Organization structured data · Google Search CentralOpened October 20, 2026. Supports: Page updated 8 September 2026: no required properties; place on the home page or one page describing the organization; recommended properties include name, legalName, address, logo, numberOfEmployees and sameAs.
- Self-Promotional Content Works, Until It Backfires (AI SEO Experiment) · AhrefsOpened October 20, 2026. Supports: Published 6 July 2026: 82% of new mentions of a little-known brand came from answers citing the authors' own pages, against 6% for an established brand; small test run by a vendor.