Try this right now: ask ChatGPT to recommend businesses in your category and city. “Best boutique homestays in Himachal.” “Who builds custom D2C storefronts.” “Good CA firms for startups in Pune.”
Someone gets named in that answer. Customers act on it — increasingly without ever seeing a search results page. If the name isn’t yours, this post is the fix, in order of impact.
How assistants actually pick who to recommend
There’s no ranking algorithm to game. A model recommends what it can retrieve and trust: businesses whose facts appear clearly and consistently across the sources it reads — your website, your structured data, review platforms, directories, articles that mention you. Three properties decide it:
- Legibility. Can a crawler read your site without executing JavaScript, and are your facts stated as facts?
- Consistency. Does every source agree on who you are, what you do, where you are, and since when?
- Corroboration. Do sources you don’t control — reviews, press, directories — back your claims?
Everything below serves one of those three.
Step 1: Make your site say something quotable
Models quote sentences, not vibes. Audit your homepage and service pages for lines a model could lift verbatim: “We’re passionate about excellence” gives it nothing. “Custom-coded ecommerce sites for premium D2C brands, live in 2–4 weeks, with full code ownership” gives it everything — the who, the what, how fast, and what you walk away owning.
Write the sentence you want repeated to a stranger, then put it high on the page in plain HTML.
Step 2: Hand over your facts in schema
JSON-LD structured data — Organization, LocalBusiness or Service, FAQPage, Article — is your facts in a format that can’t be misread: name, services, area served, founding date, contact, links to your profiles. The FAQ part matters double, because question-and-answer is the exact shape of what users ask assistants. Answer the questions your customers actually type, in one or two self-contained sentences each.
Step 3: Ship llms.txt and open the door
Two files at your domain root:
- llms.txt — a curated plain-text brief of your business: who you are, what you offer, proof, links. Cheap and worth shipping, but don’t expect much from it: Google has confirmed its search systems ignore the file, and it shows no measurable effect on citations at scale. The crawler access below is the part of this step that matters. (The full evidence, and what does work.)
- robots.txt — check you’re not blocking GPTBot, ClaudeBot, PerplexityBot and friends. Plenty of businesses block AI crawlers with a copy-pasted robots file, then wonder why they’re invisible to AI.
Step 4: Get corroborated where models read
Models weight sources you can’t edit. The compounding moves: a complete Google Business Profile with real reviews, consistent listings in the directories of your industry, and any genuine press or guest content that states what you do. One accurate third-party paragraph about your business is worth more than another page of self-description.
Step 5: Test monthly, and correct the record
Ask ChatGPT, Claude, Gemini and Perplexity your category questions every month. Three outcomes:
- You’re absent → work steps 1–4 harder; check your analytics for chatgpt.com and perplexity.ai referrers to see if it’s shifting.
- You’re present but wrong → find the stale source it’s reading (old directory, dead page) and fix it at the origin.
- You’re present and accurate → you now own a channel your competitors don’t know exists. Keep the facts fresh.
Run the loop monthly rather than once. The answers move as the models refresh their sources, and the point is to catch a stale or wrong description early — while it’s still a fix at the origin rather than a story that has spread.
I’m Divyansh Sood. Getting businesses recommended by AI is literally a service I sell — SEO and GEO, implemented in your codebase — and the deeper theory is in GEO vs SEO. Before you email me, ask the assistants about your category. Then send me what they said — that transcript is the audit.