People are asking AI about your business

Someone who would once have typed “web designer near me” into Google now asks an assistant, in a sentence, and gets an answer with three names in it. No ten blue links. No page two. Three names.

If you are not one of them, you did not lose a ranking. You were never in the conversation.

This is not a prediction and it is not a reason to panic — search is not dead and the ordinary work still matters. But there is a small set of things that decide whether an AI assistant can see you at all, and most sites have at least one of them wrong without knowing.

First: check that you are not blocking the wrong robots

This is the one almost nobody has looked at, and it is a single file.

AI companies run more than one crawler, for different purposes, and they are not interchangeable. Using OpenAI’s as the example, because they document it clearly:

• GPTBot — collects content for training. Blocking it keeps your content out of future model training.
• OAI-SearchBot — powers search. This is the one that decides whether you can appear as a source in an answer. OpenAI’s own guidance notes that publishers who allow it can see the referral traffic in their analytics.
• ChatGPT-User — fetches a page when a user’s question requires it, live.

The distinction matters enormously, and it is where the mistake happens. Plenty of site owners read an article about AI scraping, added a blanket block, and are now invisible in AI answers — while assuming they merely opted out of training. Those are different decisions and they deserve to be made separately.

The same shape applies elsewhere. Google’s Google-Extended controls whether your content helps Gemini; it does not control Google Search indexing, which is ordinary Googlebot. Blocking one is not blocking the other.

Go and look at yours. Type yourdomain.com/robots.txt into a browser. If you cannot tell what it is doing, that itself is the finding — and note that if you use a CDN, it may be injecting rules you never wrote. Ours does; we checked.

A defensible position, and the one we run on this site: allow the search and citation crawlers, block the training ones. You keep the chance of being recommended and you decline to donate your writing to a training set. Reasonable people choose differently, and either is fine — as long as it is chosen.

Second: be a source worth citing

Once a crawler can reach you, the question becomes whether an assistant has any reason to quote you. What gets picked has a consistent shape, and it is not mysterious:

Facts stated plainly, in text. Your prices, your service area, your hours, what you do and do not do — as sentences on the page, not inside an image, a PDF, or a carousel that needs JavaScript. If the fact is not extractable, it cannot be quoted.

A page that answers one question completely. Assistants pull from pages that resolve a question, not pages that circle it and invite you to enquire. This is the same shift as featured snippets, further along: the page that gets cited is the one that gave the answer away.

Being confirmable somewhere else. An assistant weighting two sources will prefer the one corroborated elsewhere — a directory, a professional body, a mention in something written by a third party. This is the old citation logic and it has not been repealed.

Consistency across the web. Your name, address and phone number identical everywhere. Where they conflict, an assistant has to pick and may pick wrong.

Being current. A page whose newest date is 2023 loses to one from this year, given the choice.

Notice how much of that is ordinary good practice. There is no separate discipline here; there is the same work, done for a reader that cannot infer what you meant.

Third: structure, so it is not guessing

Structured data is a way of telling a machine what a thing on the page is rather than hoping it works it out. It has been useful for search for years and it is more useful now.

For a local business the ones that pay: LocalBusiness with the address, phone and opening hours; Service for what you sell; FAQPage for the questions you answer; BreadcrumbList for where a page sits.

One warning, and it is not a small one. Do not put anything in structured data that is not true on the page. Invented review markup is a policy violation with real consequences, and we have written elsewhere about removing exactly that from this site — four testimonial cards with real names and words nobody said. If an assistant repeats a claim you fabricated, it will be repeated with your name attached.

What to actually check this week

1. Read your robots.txt. Decide, separately, about training crawlers and search crawlers. If your CDN injects rules, find out what they are.
2. Look for your key facts in the page text. Price, area served, hours. If they are only in an image or a PDF, move them into text.
3. Ask an assistant about your own business, by name and then by category — “who does web development in Karachi”. Read what comes back. If it is wrong, that is what your customers are being told, and it is usually traceable to something inconsistent online.
4. Check your structured data with Google’s Rich Results Test, and check that everything in it is also true on the page.
5. Keep doing the ordinary work. Most people still use search engines. This is an additional front, not a replacement.

The part worth being calm about

Nobody outside these companies knows precisely how sources are chosen, the mechanisms change, and anyone selling you a guaranteed method for appearing in AI answers is selling a guess with a price on it.

What is knowable is the floor: be reachable, be extractable, be consistent, be corroborated, be true. Every one of those is checkable today, none requires a new budget line, and all of them were good ideas before any of this.

---

Half of an SEO engagement is this kind of unglamorous audit — what is actually on the page, what the crawlers can reach, whether the structured data matches reality. If you want to know what an assistant currently says about your business and why, send us the domain.

Sources: OpenAI’s crawler documentation · the same list on the platform docs