How to get recommended by ChatGPT: GEO for small businesses

How to get recommended by ChatGPT, Claude and Perplexity: where AI assistants find businesses, a practical GEO checklist for your site and listings, and how to test yourself.

Some customers now ask an AI assistant instead of a search engine: “where can I get my sofa cleaned in Almaty”, “which nail studio near the metro should I pick”. The assistant replies with a short list of companies, and if you're not on it, the customer may never hear of you. GEO (generative engine optimization) is the work of making sure assistants know about you and describe you correctly. The good news: almost everything it takes also helps with ordinary search.

How AI assistants find businesses

Simplified, an assistant has three sources:

• Web search while it answers. ChatGPT, Claude, Perplexity, Gemini and Copilot can search the web: they send a query to a search engine, open several pages and summarise them. If your site is hard to find in ordinary search, it's harder for the assistant too.
• What others say about you: your listings on Google, Apple Maps, Yelp, 2GIS or Yandex Maps, reviews, directories, “best … in town” roundups, local news. A company described the same way in several places is easier and safer to name.
• The model's own knowledge from training. It changes only with new model versions, and you can't influence it directly, so work on the first two.

The takeaway: an assistant needs clear, consistent and verifiable facts about you in several places.

A “who we are, where, what it costs” page

Your site needs a page that states the essentials without marketing words: the name, what you do, which city and areas you serve, the address and opening hours, prices or at least “from” prices, how to order, contacts. An assistant takes facts from a page, not mood. “In-home upholstery cleaning in Almaty, all districts, sofas from [price] ₸” gives it something to quote. “We make the world cleaner” doesn't.

• Prices, address and hours as text, not images: a crawler may not read a picture.
• The important text is in the page's HTML from the start: some crawlers don't run JavaScript.
• Your robots.txt doesn't block AI search crawlers such as OpenAI's OAI-SearchBot and Perplexity's PerplexityBot.
• The site is added to Google Search Console and Bing Webmaster Tools (and Yandex Webmaster if you work in the CIS): some assistants search through ordinary search engines.

The same name, address and phone everywhere

Check how your business appears on Google Business Profile, Bing Places, Apple Business Connect, Yelp, 2GIS, Yandex Business, social networks, marketplaces and industry directories. The name is spelled the same way, the address uses one format, the phone number is the same. If you're “Clean Home” in one place and “CleanHome KZ” with another number in the next, it's harder for an assistant to tell it's one company.

• Make a table: platform, link, name, address, phone, hours, date checked.
• Fix the mismatches, starting with the platforms that bring you the most customers.
• Merge or close old duplicate listings with each platform's own tools.
• Link your site and your official profiles to each other.

An FAQ with direct answers

Collect the questions customers ask in messengers and on the phone, and answer each one on the site. One rule: the answer goes in the first sentence, the details after. An assistant can then lift a ready sentence, and a person finds the answer fast.

• Bad: “Prices depend on many factors, please contact us.”
• Good: “Cleaning a three-seater sofa starts at [price] ₸. The technician gives the exact price on site before starting: it depends on the fabric and how dirty it is.”
• One question per block, worded the way a customer asks it, not the way that suits the company.
• Include the awkward questions too: timing, guarantees, what happens if you don't like the result.

Structured data and llms.txt

Structured data — schema.org markup, usually in JSON-LD — describes your business in a language crawlers understand: business type, name, address, phone, opening hours, prices, links to profiles. Businesses usually use Organization or LocalBusiness, products use Product, a Q&A block uses FAQPage. The main rule: the markup matches what's visible on the page. If the markup shows one price and the page another, it misleads both people and machines.

llms.txt is a text file at the root of your site with a short description of the business and links to the main pages, written for language models. It's a proposed format, not an established standard, and not every AI service reads it. Add it as an extra once everything else is in order, not instead of it.

Real reviews and mentions on trusted sites

Assistants read the text of reviews, not just the stars. If customers write “came the same day” or “they cover the Bostandyk district”, those words may end up in an answer about urgent visits or that district. Ask real customers for reviews after the job, reply to them, and never buy or write them yourself: platforms fight fake reviews, and lost trust is slow to rebuild.

Mentions on trusted sites confirm that the company exists and does what it says. Where to look:

• Local media and city portals: an expert comment, news of an opening or a new service.
• Industry directories, associations, trade shows.
• “Where to find … in town” roundups: write to the author if you genuinely fit the topic.
• Partners and suppliers with “our clients” or “where to buy” pages.
• Industry forums, Q&A sites, Reddit, YouTube: answer helpfully, don't advertise.

How to check yourself in ChatGPT, Claude and Perplexity

Once a month, ask the assistants the questions your customers ask. Not “tell me about company X”, but the way a stranger to your business would ask.

• “Where can I get [service] in [city]?” and “Recommend a [service] near [area].”
• “How much does [service] cost in [city]?”
• “[Your name] reviews” and “[Your name] or [competitor]: which should I choose?”
• Ask in a new chat with search on, several times: answers vary between runs.
• Log it: question, assistant, whether you were named, what facts were given and whether they're right, which sources were cited.

The sources are the most valuable part. Perplexity, and ChatGPT and Claude with search, show links. If an assistant cites an old directory with a wrong phone number, fix the directory. If it names competitors, open their sources and see where they're mentioned and you aren't. That's your to-do list for next month.

How VIT TEAM does it

In the SEO & website content factory this is the SEO specialist's job. It writes direct-answer paragraphs for common questions, gathers the facts about the business for the “who we are” page, prepares the markup and llms.txt, and lists the maps and directories where your name, address and phone must match, while the content maker writes the page texts. For fresh data on competitors and search results, the specialist searches Google, reads the sites and brings back facts with links. The keyword map lives in its personal notebook, so new pages don't compete with old ones for the same query.

You can hire the SEO specialist for a 24-hour shift for $39 or for a month for $299. And if you already test yourself in ChatGPT or Claude, you can connect the team right there through the connector and give tasks from the same chat. You can try it free for 7 days.