AI can read a company's go-to-market from public sources: its real competitors, what customers say in reviews, who its marketing speaks to, how much search demand it captures, and whether AI assistants recommend it. Those signals cannot show revenue, retention or pipeline. A person must draw the conclusions, labelled as outside-in, and confirm them with the founder.
This is general information, not investment advice.
Key takeaways
- Public sources reveal competitors, customer voice, positioning, search demand, AI visibility and site quality.
- They cannot show revenue, retention, pipeline, conversion or what paid search actually returns.
- According to Google, Keyword Planner volumes are estimates, and very low-volume keywords are not shown at all.
- A diligence read should leave no trace: tracking blocked, no forms submitted, no cookie banners accepted.
- A person draws the conclusions, labels each as a public signal, and confirms it with the founder later.
What can AI read about a company's growth from the public web?
Six things, each from sources anyone can reach: the real competitor set, what customers praise and complain about, who the marketing is written for, how much buyer search demand exists and how much the company wins, whether AI assistants recommend it, and whether its site is built to be found.
| Question | Public sources | What it suggests |
|---|---|---|
| Who do they compete with? | Search results, review sites, buyers' own comparisons | The real set, rather than the pitch deck's |
| What do customers say? | Public reviews, forums, case studies | Whether the story matches the experience |
| Who is the marketing for? | Site, job ads, case studies, reviews | Whether it speaks to the actual buyer |
| How much demand do they win? | Keyword estimates and current rankings | The size of the organic gap |
| Do AI assistants name them? | Buyer questions put to several assistants | Visibility where buyers now ask |
| Is the site built to be found? | Technical checks and real speed measurements | Foundations for growth |
AI makes this practical in days rather than weeks, because it can read hundreds of reviews, pages and results and sort them into themes. The method for reviews is set out in the guide to competitor review research. Some sources, such as Reddit threads and full review pages, still need a person to save them.
What can public signals not show?
They cannot show revenue, margins, retention, pipeline, conversion rates or what the company's ad spend returns. They show demand, positioning and perception from the outside. A company with weak public signals may sell well through referrals. Equally, one with strong signals may struggle to convert.
So treat every outside-in score as a starting read, not a finding. The useful output is a set of questions for the founder: why do reviews praise something the website never mentions, or why does the company rank for searches its buyers do not make? Those questions often reveal more than the answers.
The company's own systems close the gap. With the founder's agreement, read-only access to analytics, the ad account and the CRM replaces estimates with real figures. Those readings then become a baseline the company can be measured against after investment.
How reliable are search estimates, reviews and AI answers?
Each is useful and each is biased. Search volumes are modelled estimates, reviews are self-selected and sometimes manipulated, and AI assistants' answers shift with phrasing and over time. Use them to compare and to raise questions, never as precise measurements, and say which is which in the read.
Search volumes first. According to Google, Keyword Planner lets you "View estimates on the number of searches a keyword gets each month", and keywords with very low search volumes are not discoverable at all. Niche B2B categories often sit exactly there.
Reviews need the same care. The ACCC warns that omitting negative reviews can mislead as much as posting fake ones, which tells you curated review pages exist. Read across several sites, and weight detailed reviews over star ratings.
AI answers vary too. According to University of Toronto researchers, AI search engines differ in freshness and domain diversity and are sensitive to how a question is phrased. So ask the same buyer questions several ways, across several assistants, and report a range. The guide to measuring AI visibility explains how.
How do you research a company without leaving a trace?
Block the company's analytics, ad pixels and CRM tags from the first page load, never accept a cookie banner, and never fill in a form, start a trial or ask for a sales meeting. Collect company-level information, not personal data about individuals. The company should see nothing in its dashboards or its sales pipeline.
A scripted browser that loads pages with tracking live adds fake visitors to the target's analytics and can feed its ad platforms. A submitted form creates a fake lead in a salesperson's queue. Both are avoidable, and the guide to testing without polluting analytics shows how.
Privacy law still applies to public pages. According to a 2023 joint statement by the OAIC and international counterparts, "Personal information that is publicly accessible is still subject to data protection and privacy laws in most jurisdictions." Diligence of a company's go-to-market rarely needs anyone's personal details, so leave them out.
Who should draw the conclusions?
A person, every time. AI gathers, sorts and drafts; an analyst or partner decides what the signals mean, how much weight they carry and what to ask the founder. Every finding should be labelled as drawn from public signals, and every score built from outside treated as provisional until the founder confirms it.
Keep the scope honest as well. Outside-in go-to-market research is not financial, legal or technical diligence. It answers questions about demand, positioning and presence, and it should say clearly what it could not see.
AI can also be confidently wrong about a company, especially a small one with little written about it. Check every claim in a draft against the source it came from before it reaches an investment committee.
What happens to the read after the deal?
It becomes the first draft of the company's own foundations: competitors, customer voice, buyer profile and positioning, all marked as drafts from public sources. The founder's own material then refines them, and a discovery call confirms them. Nothing at draft maturity goes into live work.
That continuity is the main argument for doing diligence this way. The research is not thrown away when the term sheet is signed; it becomes day one of the company's growth plan.
Skein builds Company OS systems that start from public sources at draft maturity and are confirmed by the business before anything is built on them, with every connection read only; the Investors page shows what a growth read covers. That suits funds assessing B2B companies of 10 to 200 people. It does not suit anyone looking for financial, legal or technical diligence, or a substitute for a partner's judgement.
Frequently asked questions
What is AI due diligence?
AI-assisted due diligence uses AI to gather and sort public information about a company, such as competitors, reviews, search demand, AI visibility and site quality, far faster than a person could alone. An analyst then interprets the signals, labels them as outside-in, and confirms them with the founder before anyone relies on them.
What can public data not tell you about a company?
Public data cannot show revenue, margins, retention, pipeline, conversion rates or the return on ad spend. It shows demand, positioning and how customers perceive the company. Treat outside-in scores as a starting read and confirm them with the founder or through read-only access to the company's own systems.
Are search volume estimates accurate?
They are estimates, not counts. Google describes Keyword Planner figures as estimates, and very low-volume keywords are not shown at all, which affects niche B2B categories. Use search volumes to compare opportunities and size gaps roughly, not as precise measurements, and state the source and date of every figure.
Will a company know it is being researched?
It should not. Block analytics, ad pixels and CRM tags from the first page load, never accept a cookie banner, and never submit a form, start a trial or ask for a sales meeting. Collect company-level information only, so nothing appears in the company's dashboards, sales pipeline or ad audiences.
Can AI replace an analyst in due diligence?
No. AI gathers, sorts and drafts, which saves days of reading. A person decides what the signals mean, how much they matter and what to ask the founder. AI can be confidently wrong about small companies with little written about them, so every claim should be checked against its source.
Related reading
- Google Ads Help. "Use Keyword Planner." Google, accessed 30 October 2026. https://support.google.com/google-ads/answer/7337243
- ACCC. "Online reviews for products and services." Australian Competition and Consumer Commission, accessed 30 October 2026. https://www.accc.gov.au/business/advertising-and-promotions/online-product-and-service-reviews
- Chen, Mahe, Xiaoxuan Wang, Kaiwen Chen and Nick Koudas. "Generative Engine Optimization: How to Dominate AI Search." arXiv, 10 September 2025. https://arxiv.org/abs/2509.08919
- Office of the Australian Information Commissioner. "Global expectations of social media platforms and other sites to safeguard against unlawful data scraping." OAIC, 24 August 2023. https://www.oaic.gov.au/news/media-centre/global-expectations-of-social-media-platforms-and-other-sites-to-safeguard-against-unlawful-data-scraping