How to get cited by AI assistants: a practical GEO guide

To get cited by AI assistants, make your pages reachable by their crawlers, answer specific buyer questions in the first lines of each section, and earn mentions on sites those assistants already trust. Then track the same set of questions every month. This practice is often called generative engine optimization, or GEO.

The rest of this guide turns that into steps you can run this week, with the trade-offs explained.

How AI assistants choose what to cite

Assistants like ChatGPT, Perplexity, Gemini and Copilot answer in two broad ways. Some answers come from what the model learned during training. Others come from a live web search the assistant runs at question time, and those answers usually show source links.

Live search is where you have the most control. If an assistant searches the web, it picks pages that are crawlable, relevant and easy to quote. Google describes how its AI features use the same index and ranking systems as regular search in its guide to AI features and your website.

So the practical goal has two parts: be the easiest good page to quote today, and be talked about widely enough that future models already know you.

Step 1: Make sure assistants can reach your pages

Start with access, because nothing else matters if crawlers are blocked.

1. Check robots.txt. Look for rules that block AI user agents. OpenAI documents its crawlers on its bots page, including a search crawler separate from its training crawler. Allow the search ones if you want to appear in answers.
2. Check that pages render without heavy JavaScript. Many crawlers read raw HTML. If your key copy only appears after client-side rendering, some bots will see an empty page.
3. Keep a clean sitemap. List canonical URLs only. Our free sitemap generator and robots.txt generator help if you are starting from scratch.
4. Consider an llms.txt file. It is a proposed plain-text index of your most useful pages. Support varies by provider, so treat it as a low-cost extra, not a fix. Our llms.txt explainer covers the format.

Step 2: Find the questions buyers actually ask

AI answers are built around questions, so build your list from questions too. Good sources:

• Sales calls and support tickets, word for word.
• "People also ask" boxes in Google for your main keywords.
• Reddit and forum threads in your niche.
• Search Console queries that start with how, what, which, best or vs.

Group them into four types: definitions ("what is X"), comparisons ("X vs Y"), recommendations ("best X for Y") and how-to ("how to do X with Y"). A hypothetical scheduling tool for clinics, call it Slotwise, might end up with 30 questions, such as "best appointment reminder software for physiotherapists" and "how to reduce no-shows at a small clinic."

Recommendation questions matter most for revenue. They are where an assistant names products.

Step 3: Write pages that are easy to quote

Assistants pull short passages. Write so any single section can stand on its own.

• Answer first. Open each section with a 40 to 60 word direct answer, then add detail.
• Name things fully. Write "Slotwise sends SMS reminders 24 hours before an appointment" instead of "it sends them a day before." Pronouns lose meaning when a passage is quoted alone.
• Use tables for comparisons. A table of plans, features or use cases is easy to extract and hard to misread.
• Use question headings. An H2 or H3 that matches the question makes the match obvious.
• Show dates. Add a visible "updated" date and refresh facts that change, like pricing.
• Add structured data. Schema.org types such as Organization, Product, FAQPage and Article describe your page to machines. Our schema markup generator builds the JSON-LD.

If you want a full method, the answer engine optimization skill walks through it section by section.

Step 4: Earn mentions on sources assistants trust

When an assistant recommends products, it often draws on third-party pages: comparison articles, review sites, community threads and documentation. Your own site is only one input.

Practical ways to show up there:

• Comparison and "best of" lists. Find the articles that already rank for your recommendation questions. Offer the author accurate details, a demo account or a correction if your listing is out of date.
• Community answers. Helpful replies on Reddit, Hacker News and niche forums get indexed and quoted. Read our guide on Reddit marketing without getting banned before you start.
• Directories and integration pages. Partner marketplaces and integration listings are often treated as reliable facts about what you do.
• Consistent descriptions. Use the same one-line description of your product everywhere. Mixed messages produce mixed answers.

Step 5: Track your AI visibility

You cannot improve what you do not measure, and AI answers vary from run to run. Set up a simple routine.

1. Pick 20 to 30 questions from Step 2.
2. Ask each one in ChatGPT, Perplexity, Gemini and Copilot once a month, in a fresh session.
3. Record three things: whether you are mentioned, whether you are linked, and which competitors appear.
4. Note the cited sources. Those pages are your outreach list.

Over a few months this table tells you which pages and which outside sources move the needle. A GEO audit formalizes the same process.

A 30-day GEO plan for a small team

If you only have a few hours a week, run the steps in this order.

Then repeat the visibility table at the start of the next month and compare.

A note on the homepage: many sites describe themselves in abstract terms like "the platform for modern teams." An assistant reading that cannot tell what you sell. A sentence like "Slotwise is appointment scheduling software for small physiotherapy and dental clinics" gives it a category, a customer and a job in one line. Repeat that same sentence, or close to it, on your about page, your social profiles and your directory listings.

What a citable page looks like

Here is the skeleton of a page built to be quoted, using the Slotwise example.

1. H1 matching the question: "How to reduce no-shows at a small clinic."
2. Answer paragraph: a 50-word summary naming the three methods that work best.
3. One H2 per method, each opening with a two-sentence answer, then steps and a short example.
4. A comparison table of reminder channels (SMS, email, phone) with cost and effort.
5. A short FAQ with three follow-up questions, marked up with FAQPage schema.
6. A visible updated date and an author or team name.

Every section can be lifted on its own and still make sense. That is the test to apply before you publish.

Common mistakes that block citations

• Gating your best answers. Content behind a form or login cannot be fetched.
• Vague category language. If your homepage never says what category you are in, assistants cannot place you.
• Out-of-date facts. An old pricing page that contradicts the new one gives the model two answers to choose from.
• Thin "AI-friendly" pages. Short pages stuffed with questions and no substance do not hold up. Google's helpful content guidance still applies.

How Thig helps

Thig's GEO agent tracks how ChatGPT, Perplexity, Gemini and Copilot describe your category, shows which competitors and sources they cite, and turns the gaps into content and fix suggestions in your feed. The SEO agent and Coding agent handle the crawl and markup side.