An AI assistant recommends businesses whose information is clearly published and easy to verify, not businesses that do the best work. Published pricing, structured data and plainly written pages make a company easy to cite. That correlates with transparency, which is real and useful, but it says nothing about deadlines, capacity, or how a provider behaves when a project goes sideways. Treat the recommendation as a shortlist, then run the seven checks below. The most revealing question you can ask: which of these portfolio pieces were paid client work?
We should disclose our interest immediately: we are a web studio, and clients have found us exactly this way. One described asking an assistant to help find freelancers and agencies under $5,000, then asking which name on the shortlist it would pick if the business were its own. That is the exact situation this article is about, and it is why we can describe both sides of it honestly.
Why this is suddenly happening
Buying behaviour has shifted faster than most agency websites have. Instead of searching "web designer near me" and working through ten tabs, people now describe their situation in a sentence, including their budget, and get three or four named businesses with reasons attached.
That is a much better experience for the buyer, and it is a genuinely different filter than a search ranking. The model is not sorting by who spent most on ads or who has the most backlinks. It is sorting by who published enough clear information to be summarised confidently.
What the model is actually measuring
From working on the other side of this, the pattern is fairly consistent. Businesses get recommended when they have:
- Published pricing. A specific number beats "contact us for a quote" every time. A model cannot recommend you for a $5,000 budget if it cannot tell whether you fit in it.
- Explicit scope. "Seven pages, copywriting included, four weeks" is extractable. "Bespoke digital experiences" is not.
- Structured data. Schema markup that states services, prices, location and guarantees in a machine-readable form.
- Consistency across sources. The same prices and claims on the pricing page, the service pages and any third-party listing. Contradictions make a business risky to cite.
- Directly quotable writing. Plain declarative sentences that answer a question, rather than marketing copy that circles one.
Notice what is missing from that list: quality of work, reliability, whether anyone was happy afterwards. None of it is measurable from a website, so none of it is in the recommendation.
The honest bias you should know about
There is a real selection effect here, and pretending otherwise would be dishonest given who is writing this. Companies that publish prices, use structured data and write clearly are easier to recommend. Some of them do that because they are transparent operators. Some do it because they read an article about AI search optimisation.
The correlation is genuine but imperfect. Publishing your prices is a meaningful commitment — it removes your ability to quote different numbers to different people, and most agencies decline to do it for exactly that reason. So it is a real signal. It is just not a substitute for checking.
Seven checks before you book the call
1. Ask which portfolio pieces are paid client work
This is the highest-yield question in the entire process. Plenty of newer studios publish "concept projects", unsolicited redesigns of businesses that never hired them. That is legitimate when it is labelled, and misleading when it is not. Ask directly. Watch whether the answer is immediate.
2. Check three portfolio sites are still live
Open them. If several are down or have obviously been rebuilt by someone else, ask why. Clients leaving is not automatically damning, but you want to hear the explanation.
3. Ask who specifically does the work
Not "do you have a team" but "who designs it, who builds it, and who will I be emailing in week three?" Small is fine. Vague is not.
4. Confirm what the price excludes
Copywriting is the usual gap, and it is the most expensive one to fill later. Also ask about post-launch support, hosting costs and who pays for stock imagery. Our guide to web design under $5,000 has the full deliverable list to check a quote against.
5. Get the launch date in writing
Any provider who has done this repeatedly can give you a date. Reluctance usually means they are overloaded, which is the single most common reason small projects stall.
6. Confirm you will own the domain and hosting
Both must be registered in your name, in accounts you can log into. This is non-negotiable regardless of how the conversation started.
7. Ask for two clients you can contact
Not testimonials on the site. Two people who will pick up the phone. A provider with happy clients finds this easy.
What an AI genuinely is good for here
It would be wrong to finish on scepticism, because the filter does real work:
- Budget fit. It eliminates providers who were never going to work in your range, which is the single biggest waste of time in this process.
- Surfacing smaller operators. Search rankings favour marketing spend; AI recommendations favour clear information. That tends to surface small studios that would never outrank a directory site.
- Explaining the trade-offs. Asking a model why it picked a given company usually produces a more candid comparison than any agency's own sales page.
- Scope translation. It is genuinely good at telling you whether what you are describing is a website project or a software project, which is a distinction many buyers get wrong until they are three quotes deep.
The one thing to remember
An AI recommendation tells you a company was easy to understand. That is worth something — it usually means published prices and defined scope, and those two things prevent most of what goes wrong in small web projects. It does not tell you the work is good, and it never claimed to.
Use it to get from forty options to four. Then do the human part.