You get an email from a supplier you have never heard of. Good price, decent website, a name in the signature. What do you do before you answer? Ten years ago you opened Google and typed the company name plus "reviews". Today a growing share of people open ChatGPT instead and type: "Is this company legit? Any red flags?" They get a paragraph back in ten seconds, with sources, and they decide.
Here is the part most business owners have not caught up with: the same thing is happening to you. Your prospect, your next hire, the investor you emailed last week, the journalist who might quote you. Before they reply, they ask an AI assistant who you are. What it says is now the first impression, and you have never seen it.
This is our own article about our own tools, and I will say that plainly rather than pretend it is neutral. We built three small checks for exactly this moment: AI Company Scan, AI Person Scan and AI Website Visibility. This piece explains what they do, what they cost, when they are worth it, and what they cannot do. If you finish it and decide you can do the same thing by hand, that is a fine outcome too.
The shift in one table
The numbers are not subtle. Search is not dying, but the question "who is this?" is moving from a list of links to a single answer.
| What changed | The figure | Where it comes from |
|---|---|---|
| Traditional search volume | Down 25% by 2026 as chatbots take the queries, Gartner's forecast | Gartner |
| Visits AI assistants send to websites | Roughly 60 million a month in October 2024, around 240 million by September 2025 | Similarweb |
| Google searches that end without a click | 68% | Similarweb, citing Rand Fishkin's research |
| How a ChatGPT visitor behaves on a site | 15 minutes on site against 8 from Google, and a 7% conversion rate against 5% on transactional sites | Similarweb |
| Money lost to email fraud that impersonates vendors and colleagues | $2.77 billion in 2024 across 21,442 reported cases in the US alone | FBI IC3 2024 report |
| People who look someone up before meeting them | 77% of Americans surveyed by JDP have researched a date; half of Avast's dating-app users searched a match, and 30% of those decided not to meet | JDP via Bustle, Avast |
Two honest notes on that table. Gartner's 25% is a forecast, and the data through mid-2026 shows AI answering more questions while classic search still carries far more traffic; the pie grew rather than the old slice shrinking. And the "look someone up" numbers are about dating, because nobody has published a clean survey about business meetings yet. I would bet the business figure is higher, not lower: nobody wires €40,000 to a stranger with less research than they put into a first drink.
The point holds either way. The assistant's answer is now a step in the buying process, and it is a step that runs without you in the room.
Why checking with AI beats not checking at all
Most small and mid-sized companies have no security department. They have a founder, a few people, and a rule of thumb: if the website looks fine and the email is polite, go ahead. That rule is what the $2.77 billion in the FBI's table is built on. Vendor impersonation works because checking is a chore, so people skip it.
An AI assistant with web search turns the chore into a ninety-second habit. Ask it "who is behind acme-supplies.com, what do they sell, and are there complaints, lawsuits or scam warnings?" and it reads the company's site, the registries and review platforms it can reach, the news, the forums, and comes back with a summary and links. It is not an investigator. It will miss things a paid due diligence firm finds, and it will sometimes confuse two companies with the same name. But the red flags that matter most for a small business, a trail of unpaid contractors, a regulator's warning, a domain registered last month, a founder with a previous company that folded owing money, show up in exactly the places these assistants read.
The trade is simple: a formal vendor due diligence report costs hundreds to thousands of dollars and takes days; a security team costs a salary. A five-model AI check costs less than lunch and takes five minutes, and for the deals under €50,000 that make up most of a small company's year, it is the difference between checking and not checking.
Three questions, three tools
We built one check per question people actually ask. Each one starts free, so you see what the model finds before you pay anything.
"Is this company safe to work with?" AI Company Scan
You enter the company's website. One model runs a free preview and tells you which company it thinks is behind the address, so you pay only if it found the right one. Then five AI models with live web search, OpenAI, Anthropic, Gemini, Perplexity and Grok, each answer the same three questions: who is behind it, what does it sell and to whom, and are there red flags about working with it. The report is one PDF: who they are, what they do, red flags with sources, where the models disagree, other companies with the same name, and every answer word for word. €14.97 once.
This is the vendor due diligence step for the deals that never justified a real one: a new supplier, an agency, a distributor, a partner who found you on LinkedIn.
"Who is this person?" AI Person Scan
You paste a LinkedIn, X, Instagram or Facebook profile link. The free preview shows the name, role and company the model finds, so a namesake does not slip through. Then the same five models answer who the person is, what they do professionally, and whether there are red flags about working with them. You get their background as AI tells it, their latest public work with dates, disputes and reviews with sources, the contradictions between models, and the namesakes they got mixed up with. €14.97 once.
Most people run it on themselves first. That is the right instinct: the Person Scan is the answer your next client sees before they reply. We published what happened when five models were given one founder's name and profile in a joint paper with ANDEKS, Right, but Unproven: one model found nothing it could tie to the person at all, and the ones that answered agreed on the facts but could not prove most of them from public pages.
"Can AI read my site?" AI Website Visibility
This one is not about reputation. It is for the owner of the site and the developer who can change it. The free check reads your homepage the way a crawler does and scores eight technical signs: whether robots.txt lets the answer-engine bots in, what a bot sees without JavaScript, structured data, citation readiness, structure, first response time, HTTPS, and whether an llms.txt exists. Then, in the paid part, the five models look your site up with live search and the report shows what each one found and whether it cited your pages or somebody else's, followed by the fixes in order, with prompts and templates for your developer. AI Fix Kit, €49 once, and a code for 50% off up to seven more checks.
Why it exists: on 29 September we ran it on a partner's site, super.tennis, a tennis site with a 92 out of 100 technical score. Four of the five models cited its pages. Gemini described a different site entirely, the Italian TV channel supertennis.tv, and Claude mixed in a Nintendo game from Wikipedia. A perfect ranking in Google would never have shown that. You can read the full sample report before you run your own.
Why one SEO ranking no longer gets you into the answer
Google and AI crawlers read sites differently, and this catches good sites all the time. A robots.txt rule written years ago can let Googlebot in and turn OAI-SearchBot and Claude-SearchBot away. Text that loads through JavaScript reaches a browser and Google, but most AI crawlers do not run scripts, so to them the page is empty. And an assistant with search does not rank pages, it picks a handful to read and quotes them, which is why a competitor's blog post about you can become "the source" while your own site is skipped. I wrote about the mechanics, and why so many experts are missing from the answer to "who can help with AI governance", in Your Name Is Not in the Answer.
None of this is a promise that fixing the site gets you recommended. Nobody can promise that, and anyone selling it is selling something else. Readability is the first step, not the last. But a site the assistants cannot read is out of the answer by definition, and that is the part you can fix this week.
The money, honestly
Here is the arithmetic for a small company, using round numbers you can replace with your own.
- A bad vendor. A €20,000 order with a supplier who does not deliver, or a €3,000 monthly agency retainer that produces nothing for six months, is a five-figure hole. A €14.97 check that surfaces one regulator's warning or a founder's previous collapse pays for itself roughly 1,300 times over on the first save. You will run many checks that find nothing. That is the point of a check.
- A lost prospect. If ten prospects a month ask an assistant about you and one walks away because the answer is thin, wrong, or about a namesake, and your average deal is €5,000, that is €60,000 a year lost to a paragraph you never read. Reading it costs €14.97.
- An unreadable site. Similarweb's data says a visitor arriving from ChatGPT converts at 7% against 5% from Google and stays almost twice as long. Those visitors only arrive at sites the assistants can read and cite. If your site is not one of them, the €49 kit is the cheapest fix on your marketing list, and the free check tells you in a minute whether you need it.
Compare that with the alternatives. AI visibility monitoring subscriptions start at $95 to $99 a month and track how often you are mentioned; they do not check a counterparty and they do not write the fixes. An agency GEO audit is quoted from $1,500. A formal background or due diligence report is hundreds of dollars per subject. Our three checks are built for the volume end of the market: many small decisions, each too small for a professional and too big to leave to luck.
What this means for you
If you buy from small suppliers. Make the Company Scan part of the onboarding form. Website address in, report by email in about five minutes, one line in your notes: "checked, no flags" or "checked, one dispute, asked about it." It is the same habit as asking for a VAT number, and it catches what a VAT number does not.
If you sell your own expertise. Run the Person Scan on yourself before your next outreach campaign, not after. If a model finds nothing, or finds someone else, that is what your prospects are being told, and the fix is public pages that put your name next to your role, your company and your city. The report's recommendations are specific about which pages. Then run it on the three people you are about to negotiate with.
If you run a company site. Run the free visibility check now, it takes a minute, and hand the result to whoever maintains the site. Most fixes are a developer's afternoon: a robots.txt line, server-rendered text, an Organization schema block. If you are building something new, check the landing page before you spend on traffic, because a launch nobody's assistant can read is a launch at half volume. And if you are one of the agencies and freelancers who sell to small businesses, the free check is the best cold opener we know: "ChatGPT cannot read your homepage, here is why" gets more replies than "we do AI automation."
If you are choosing a market. Before you commit to a vertical, run the incumbents through the Company Scan. Where the models surface complaints and warnings about the established players, a market's reputation is soft, and that is where a careful newcomer wins.
What these checks cannot do
We would rather you know the limits than find them in a report.
The models say what they say. We record it with its sources; we do not verify it and we do not claim it is true. Ask twice and the wording changes, sometimes the facts. Every report is a snapshot with a date on it. The answers come from each provider's API with search on, not from the consumer apps, which add their own instructions and remember their users, so your own ChatGPT may say something slightly different.
The Person Scan is not a background check, and we are not a consumer reporting agency: do not use it for hiring, tenancy, credit or insurance decisions. It shows what AI says about a person, which can be wrong or about someone else, and the report says so on its cover. The Company Scan checks the organisation behind a website, not its legal ownership or registration; for a registry extract you still need the registry. And the visibility check reads the homepage, robots.txt and llms.txt on a live request, the way a crawler does; it does not crawl deep pages, and it does not measure whether you will appear in anyone's answer.
If you run a customer-facing AI agent yourself and want it checked the way a customer would, that is a different product, the AI Test Purchase, and a different article. We tested twelve US state DMV chatbots with the same ten questions this month if you want to see what that looks like.
The honest take
The economy is turning agentic in a very ordinary way: not robots negotiating with robots yet, but people asking an assistant before they ask a person. Every counterparty you have is a paragraph in a chat window now, and so are you. The companies that adapt will do two cheap things. They will check before they sign, because checking finally costs less than the risk. And they will make sure the paragraph about them exists and is right, because the paragraph is being written whether they read it or not.
Start with the free previews: check a company, check a person, check a site. Then decide whether the full reports are worth €14.97 and €49 to you. Most of what we have said above is verifiable in the sample reports on those pages, which is how it should be for a product that sells verification.
