How ChatGPT, Perplexity and Google AI Overviews decide whether to mention your business

Something changed in the last two years and most Indian businesses haven’t noticed yet. A good share of buyers now open ChatGPT, Gemini or Perplexity before they open Google. They ask a question in plain language and they get three or four company names back.

If your company isn’t one of those names, you never find out. There’s no ranking report for this. Nobody sends you a notification saying you were left out.

These systems read text, and only text

This is the whole thing in one line. An assistant reads the words on your page. It can’t read a scanned PDF catalogue. It can’t read a price list saved as a JPEG. It can’t read the specification table your designer exported as an image because it looked neater that way.

We audit a lot of sites where ninety percent of what the company actually knows is locked inside images and PDFs. To a search engine that content is close to invisible. To an AI assistant it may as well not exist.

The fix is unglamorous. Take the specifications out of the image and put them on the page as text. That’s it. It takes a few weeks for a large range and it’s the highest return work in most rebuilds.

They quote sources they can verify

Assistants that cite sources prefer pages that state things plainly and back them up. A page that says we’re a leading manufacturer gives them nothing to work with. A page that says we make MS ERW pipes from 15mm to 150mm NB, IS 1239 certified, from a plant in Sanand gives them four specific facts they can repeat.

Vague marketing copy is a real liability here in a way it never was before. Search engines tolerated it. Assistants skip past it because there’s nothing quotable in it.

What they actually look for

• Specifics with numbers. Sizes, grades, tolerances, capacities, turnaround times, certifications with their actual numbers.
• Location, written out. City and state written out as text on the page. A map embed on its own doesn’t count. A buyer asking for suppliers near Coimbatore needs the word Coimbatore to exist somewhere.
• Questions answered directly. A heading that asks what a buyer asks, followed by a straight answer in the first two sentences.
• Structured data. Schema markup tells a machine what kind of thing your page describes. It’s invisible to visitors and it removes guesswork for everything else.
• A page per thing. One page covering forty products gives an assistant one blurry signal. Forty pages give it forty clear ones.

The llms.txt file

There’s an emerging convention called llms.txt. It’s a plain text file at the root of your site that summarises what your business does and links to your main pages. Think of it as a briefing note written for a machine rather than a person.

It isn’t an official standard yet and no assistant is obliged to read it. It costs about an hour to write. We put one on every site we build because the downside is an hour and the upside is being described accurately.

Where businesses get this wrong

The common mistake is treating this as a new marketing channel that needs new spending. It mostly isn’t. Nearly everything that makes you legible to an AI assistant is the same work that makes you legible to Google and to a human buyer on a phone.

Publish what you know. Write it as text. Be specific. Say where you are. The businesses that already did that for search engines are the ones getting named by assistants now, and they didn’t do anything extra.

How to check where you stand

Open an assistant and ask it the question your buyer would ask. Best pipe suppliers in Gujarat. Pathology labs in Pune that do thyroid tests. NABL accredited labs near me. See who gets named.

Then ask it directly about your company by name and see what it says. The same exercise, done properly, is in how to check what an AI assistant says about your business. Sometimes the answer is worse than being left out. We’ve seen assistants describe companies using details that were correct in 2012, because that’s the newest text they could find.

Written by Sameer Gaikwad, founder of Buildgin. Thirty three years in electrical power systems across global MNCs, now applying the same condition monitoring habit to websites.