Google is no longer the only place people search
Customers now ask ChatGPT and other assistants for recommendations. What makes a site readable to them, and why it overlaps with good SEO.
A customer looking for a supplier used to type a few words into Google and work down the results. A growing number of them now ask an assistant instead — ChatGPT, Perplexity, Gemini, or the AI answer that sits above Google's own results — and get back three recommendations with a sentence of reasoning each. Whether your business is one of the three is decided by things you can control, and most sites get them wrong.
What is actually happening when someone asks an assistant
There is no mystery ranking spirit. Broadly, two things are going on. The assistant draws on what it absorbed during training — which is where long-standing, widely-mentioned businesses have an advantage — and it fetches live pages, reads them as text, and quotes what it can verify.
That second part is where a small business can win, and it has a hard requirement: the page has to be readable. Crawlers that gather pages for AI assistants generally do not run JavaScript. They request the URL, take the HTML that comes back, and move on. If your headline, services and prices only appear after a framework boots in the browser, the assistant sees an almost empty document and recommends somebody else.
This is the single most common reason a perfectly good-looking site is invisible to AI search. It is also the easiest thing to test: view the page source and search for your own services. If they are not in there as text, neither is your business.
What makes a page quotable
An assistant prefers claims it can lift without interpretation. Pages that get quoted tend to share a few traits:
- A direct answer near the top. If the question is "what does this firm do and where", the first hundred words should answer it in a sentence someone could paste into a reply.
- Specifics over adjectives. Areas served, hours, starting prices, turnaround, the exact services. "Affordable, reliable, trusted" carries no information and gets skipped; "₹10,000–₹50,000, Mumbai and remote, typically two to four weeks" can be quoted straight.
- Headings shaped like questions. An FAQ is not decoration; it is a block of pre-written answers in the exact shape an assistant needs.
- One stable name for the business, used identically on the site, the Google Business Profile, LinkedIn and every directory. Assistants resolve entities; three spellings of your name split you into three weaker ones.
- Structured data. Schema.org JSON-LD for the organisation, the services, the FAQ. It is a second, unambiguous copy of the same facts.
- Corroboration elsewhere. The same details on other sites — profiles, listings, write-ups — raise confidence. A claim that appears in exactly one place is a claim with one source.
A quick self-test
Ask an assistant the question a customer would ask: "Who does commercial kitchen repair in Andheri?" Then ask it, plainly, what it knows about your business by name. The first answer tells you whether you are in the running. The second tells you what it has actually read — and it is usually a short, slightly out-of-date list that reveals exactly which facts are missing from the open web.
This is not a separate discipline from SEO
Nearly everything above is good traditional SEO: readable HTML, clear titles, accurate structured data, consistent business details, real content that answers real questions. Answer-engine optimisation is mostly the same work with the emphasis moved — away from keyword placement and towards being quotable and verifiable.
Two things genuinely differ. First, assistants summarise rather than list, so there is no position two consolation prize; being the source that gets quoted matters more than being the tenth link. Second, citations are unstable. The same question can produce different recommendations on different days, from different assistants, for different users. Anyone selling you guaranteed AI visibility is selling something that does not exist.
What you cannot control, and should budget for
- Caching and lag. A change on your site may take weeks to show up in an assistant's answer, and some of what it says will be out of date in the meantime.
- Hallucinated details. Assistants occasionally invent prices or hours. The defence is publishing the real ones prominently, in text, in more than one place.
- Where the traffic goes. An assistant that answers the question may mean fewer clicks to your site even as more people hear your name. Enquiries, not sessions, are the number to watch.
The short checklist
- Your services and prices exist as text in the delivered HTML
- The first hundred words answer who you are, what you do and where
- An FAQ written in the customer's own words
- Organisation, service and FAQ structured data in place and accurate
- One spelling of your name, one address, one phone number, everywhere
- Prices or price bands published, not hidden behind an enquiry
- robots.txt that does not accidentally block AI crawlers you want
What we would do
We run the pages you care about, show you what a crawler actually receives, and fix the gap between that and what a visitor sees. Then the quotable parts get written: the direct answer at the top, the FAQ, the structured data, the consistent details across your profiles. That work sits in the ₹10,000–₹50,000 band, and we will tell you before starting if we think your site needs rebuilding rather than patching.
More from the build
- Websites What a business website should do in its first 90 days Traffic is not the first goal. Enquiries are. What to put on the page, what to leave off, and how to tell whether it is working.
- Automation The WhatsApp-to-spreadsheet problem Why so many businesses run on chat plus a sheet, what it quietly costs, and the smallest change that ends the retyping.
- Software Before you build an internal tool, map one job end to end Most failed internal software was scoped from an org chart instead of a job. One afternoon with a pen fixes that.
- AI What an AI pilot should cost, and how to tell if it worked Pick one repeated task, put a number against it before you start, and run it for a month. Everything else is a demo.