# Everyone has a startup now. Being found is the hard part. Shipping is easy now. Being found is not, and buyers increasingly ask an AI before they search. What I learned from sites nobody had written about. By Dasha. Updated November 3 2026. https://heardofyou.io/blog/everyone-has-a-startup-now Everyone and their mother has a startup now. Mine included, and I mean that literally: my mum's food business runs on software I wrote for her, and as of last week I have a new company of my own. It has a website, a logo made of stained glass, a free thing and a paid thing. It has been seen by roughly eleven people, and I am related to several of them. ## Building was the easy part I have a day job in data and analytics. On the side I run a small Etsy shop, I built the ordering system for my mum's business, I have made websites for clients, and I have more unfinished side projects than I will admit to in writing. Ten years ago each of those would have cost a month of evenings. Now I can put up a decent site in an afternoon, with a form that works and a database behind it. So can everyone else. That is the whole problem in one sentence. "If you build it, they will come" was written by someone with a marketing budget, and the queue of people who can build has gone from a few founders with a technical friend to everyone with a laptop and a chat window. Shipping stopped being the hard part. Being found is. ## What it looked like from the outside From the outside my sites looked fine. They loaded fast. They said the right things. Search engines indexed them within a week. A couple of them even ranked for their own name, which is the internet's way of patting you on the head. Then I did the thing I now do for a living. I asked ChatGPT, Gemini, Claude and Perplexity the question a buyer would actually type. Not "what is this site" but "what is the best tool for the thing this site does". I asked each assistant several times, because they do not give the same answer twice. Across all of those answers, my carefully built sites came up zero times. Not last. Not "also worth a look". Zero. ## What was actually true The sites were not bad. Nobody had written about them anywhere an AI reads. Here is the mechanism, as far as I can tell from a lot of research and then my own checks. When you ask an assistant for a recommendation, it does not consult some deep understanding of your market. It searches, finds the pages that already rank for that question, and summarises who those pages like. "Best invoicing tools for freelancers" lists. Review sites. A Reddit thread where somebody asked the same thing in 2024. If you are on those pages, you might be in the answer. If you are not, you are not, and no amount of polish on your own homepage changes that. The AI did not evaluate anyone. It is high school with citations. The numbers that made me take this seriously, with who paid for each of them, because somebody always did: - 51% of B2B software buyers now start their research in an AI chatbot more often than in Google (G2 Answer Economy report, March 2026; G2 sells review visibility, so this is a number with a sales team). - Among people who use AI to research a purchase, 63.5% use Gemini and 60% use ChatGPT (Product.ai, 550 people surveyed). - In a test of 2,961 prompts, the same brand list came back twice less than one time in a hundred (SparkToro, with a tracking vendor). A screenshot of an AI naming you is evidence of one Tuesday. Put those together and the shortlist is being written before anyone visits your site, it is written from pages you do not control, and it changes every time someone asks. An assistant that sounds certain about who is best is not measuring anything. It is summarising whoever got written about. ## The rule I follow now Check, work, check again. Ask the real buyer questions, several times, on several assistants, and count how often you are named. Then do the dull work of getting onto the pages the assistants already read: the lists, the review sites, the threads, honestly and under your own name. Then ask the same questions again, on a date you set before you started, and publish the number whether or not it flatters you. I am not going to pretend I know how much of this a one-person company can move, or how fast. Nobody does. The people who say they do have confetti in their launch posts. So the plan is to find out on myself first. On Friday November 6 I run the full check on Heard of You: 27 buyer questions, three times each, on four assistants. My forecast is that I will be named in none of them, because the site is a week old and nobody has written a word about it, including me, until now. Then I do the work I sell, on myself, and publish the number again at day 30, 60 and 90. If it stays zero you will hear about that too, in a slightly smaller font. The method is written down at [how we measure](/how-we-measure), and if you would rather run it on yourself by hand, the [do-it-yourself guide](/guides/check-if-chatgpt-recommends-you) takes about an hour. If you built something good and the robots keep recommending someone else, come along. I will tell you what works, and I will be equally loud about what does not. ## Sources - G2 Answer Economy report, March 2026 (https://company.g2.com/news/g2-research-the-answer-economy) - Product.ai survey of AI shoppers (n=550) (https://finance.yahoo.com/technology/ai/articles/product-ai-survey-finds-four-120000400.html) - SparkToro, AIs are highly inconsistent when recommending brands, 2026 (consultancy with a tracking-vendor partner) (https://sparktoro.com/blog/new-research-ais-are-highly-inconsistent-when-recommending-brands-or-products-marketers-should-take-care-when-tracking-ai-visibility/)