Practical
Your Buyer Asked a Robot. You Weren't There
I decided to check out whether AI searches would find my client. I did dozens of searches and the results were surprising, and not all good. But there does appear to be a simple fix. How does your company show up?
Otto Pohl
Share this post
Last week I ran an experiment. I searched 46 times, across 5 different AI assistants, to see if my client could be discovered by AI.
I channeled the target customer and asked how to solve the problem whose solution is my client’s key value proposition. The company’s product is in market, they’ve closed serious funding, and they have better technology than anyone else in the category. There’s lots of coverage about them online. Yet they never came up in entire categories of AI searches. Not once.
Uh oh.
When I changed the question and asked "who is doing [the specific technology by name]," my client came up first, every single time. Five engines, 19 out of 19 times, named ahead of every competitor and research institute, with correct deployment sites and accurate funding figures.
Turns out, my client isn’t invisible. They’re filed under the wrong concepts. It’s fixable, but only when you know the problem exists.
A few notes before I dive in:
To keep results clean, I only used AI when I was logged out and in Chrome Incognito mode. This also means I could only use simpler models.
One surprising learning was how different the results were, even when I kept everything the same—model, query, etc. The results would look different, name different companies, cite different research, everything. There was noise everywhere except in the results I found.
The Key Issue
What seems to best explain the results is that these AI models don’t maintain one unified body of knowledge. They generate separate information trees clustered around concepts, which means facts about your company may become siloed. Different prompts trigger results from almost entirely separate areas of data. Despite the all-knowing, confident tone that AI models strike, what seems to best explain what I saw is this: once they’ve gone down a pathway and constructed a response that is internally consistent with the information they find there, they call it a day.
Three Categories, Three Outcomes
My queries targeted three areas:
Buyer questions—how can I best solve this problem?
Market questions—what startups are out there that solve this problem?
Solution questions—what companies sell this specific technology?
Query: How Can I Best Solve This Problem?
What happened: Client totally AWOL.
Why: The answers I saw were substantially assembled from vendors' own technical explainer pages, and the companies that published those webpages got the credit. One AI engine even cited a research paper that my client’s CTO had written to explain how the problem gets solved. His name (but not company affiliation) is right at the top. The model summarized the work accurately, credited it to "research groups," and then recommended another company.
What we’re doing: You don’t need coverage in tier-1 publications to solve this. My client needs articles, PDFs, and web pages that discuss their technology in terms of the problem. The documents that currently exist on the company website are behind email and account gates. So: write the explanation of your customer's problem yourself, put it on your own domain in plain HTML, don't put it behind a form, and title it after the failure mode instead of your technology. Be fair to competing approaches and mention where yours is the wrong tool. The pages sourced most often read like references, not brochures.
Query: What Startups Solve This Problem?
What happened: Mostly absent
Why: We’ve secured tons of startup coverage. But those articles hype company progress like funding, factories, and product launches, and don’t connect the problem and the company. To answer my query, the model needs documents that tie together the company's name, funding status, and the problem. My client's coverage splits them. The articles that call them a venture-backed startup don’t highlight the failure mode they fix. The articles about the failure mode don’t mention that they're a startup.
What we’re doing: I noticed a pattern in smaller competitors who kept cropping up. Where my client had coverage with headlines like "Startup raises $100M to speed up chip manufacturing", they had “Startup announces [Technology] that solves [problem].” We need more coverage that explicitly connects company, technology, and problem, so the AI models see the connection. I also noticed how often the models quoted industry overviews published by analysts. I’m going to spend more time making sure to get that coverage.
Query: What Company Sells This Specific Technology
What happened: My client crushed it, leading the AI results 19 out of 19 times.
Why: My client is the market leader and we earn lots of great coverage. The AI noticed.
What we’re doing: Popping champagne.
Run the experiment on your company
Read my full experiment report and recreate it today. Log out of everything, use a private browser window, start a fresh chat for each one, and ask four questions three times each:
How do I solve [your customer's specific problem] without [the constraint]? (Don’t name your technology.)
I'm evaluating [category] options. Who should I be looking at?
What startups are working on [category]?
Who is doing [your technology's name]?
Record which companies come up, in what order, and every source the model cites.
If you show up on question four and nowhere else, you have my client's problem. You are indexed under your technology, not your customer's problem, and no amount of additional, similar material will change it.
By the way, if the sources cited on a question are mostly junk, videos and forum posts and blogs from unrelated industries, that's not a failure, it’s an opportunity. It means that information tree is still under-populated, and you can fill it.
The background check moved
In my article “Your Website Isn’t a Billboard. It’s a Background Check” I wrote that a sales lead will visit your site looking for reasons not to buy.
Now there's a background check that runs before anyone calls you at all, conducted by a machine using an information hierarchy that might be working against you.
Go ask it about yourself, in your customer’s words. Tell me what you find.
Share this post
Otto Pohl is a communications consultant who helps startups tell their story better. He works with deep tech, health tech, and climate tech leaders looking to create profound impact with customers, partners, and investors. He has taught entrepreneurial storytelling at USC Annenberg and at accelerators across the country.
