Imagine this. You’ve spent two years building a genuinely good product page and a blog full of useful guides. Then you ask ChatGPT, “What’s the best tool for [your category]?” and it names three competitors you’ve never worried about. You aren’t mentioned. Perplexity cites a Reddit thread and a comparison site. Gemini lists someone else’s pricing page.

Your first instinct is to write more blog posts. That’s usually the wrong first move.

If you want to get cited in ChatGPT and the other AI tools, the work starts with some unglamorous checks, then moves to what you publish, and then to what other people say about you. Here’s the order I’d tackle it in, with the data and the caveats.

Mentions vs. citations (and why it matters)

These two get mixed up constantly. A mention is when an AI names or recommends your brand in its answer. A citation is when it links your page as a source. You can have one without the other.

The platforms also behave very differently:

  • Across 680 million citations analyzed in one study, only 11% of domains were cited by both ChatGPT and Perplexity.
  • A 2026 study of 34,234 AI responses found ChatGPT cited brands just 0.59% of the time, while Perplexity sat at 13.05%.
  • Perplexity cites far more sources per answer than the others, while ChatGPT is more selective.

So “optimize for AI” isn’t one job. It’s three or four small ones. Let’s go through them in order.

Step 1: Stop accidentally blocking the bots

This is the most boring step and the one I’d check first, because it’s the easiest way to make yourself invisible.

OpenAI documents separate crawlers with separate jobs. OAI-SearchBot is what surfaces websites in ChatGPT’s search features, and sites that opt out of it won’t be shown in ChatGPT search answers. GPTBot is a different bot, used for training. OpenAI says each setting is independent, so you can allow OAI-SearchBot and disallow GPTBot.

This trips people up. Blocking GPTBot doesn’t remove you from ChatGPT search, but blocking OAI-SearchBot does. A third bot, ChatGPT-User, handles user-initiated actions. OpenAI notes that robots.txt changes can take about 24 hours to be reflected.

A simple robots.txt for “show me in search, don’t use my content for training” looks like this:

User-agent: OAI-SearchBot

Allow: /

User-agent: GPTBot

Disallow: /

Decide for yourself whether you want the training block. The point is not to block the one you want to be discovered by.

Perplexity works similarly. Its documentation says PerplexityBot surfaces and links sites in Perplexity search results, isn’t used for training, and that Perplexity recommends allowing it. One vendor audit of 175 brands says the most common mistake is a default disallow on AI bots in a starter robots.txt template, which removes the URL from Perplexity’s candidate pool. Treat that as one agency’s observation, but checking your file takes two minutes.

Google/Gemini: Google’s guidance is blunt. To appear in AI Overviews or AI Mode, a page must be indexed and eligible to be shown in Google Search with a snippet, and there are no additional technical requirements. Robots.txt rules for Googlebot are the control for crawling.

Quick check: open yoursite.com/robots.txt, look for blanket “Disallow: /” lines under bots you care about, and look at your server logs for hits from OAI-SearchBot and PerplexityBot.

Step 2: Get found by the search engines underneath

These tools don’t browse the web from scratch. They lean on search indexes.

Seer Interactive’s controlled study found 87% of SearchGPT citations match Bing’s top results, while only 56% match Google’s. That’s why I’d set up Bing Webmaster Tools alongside Google Search Console and submit your sitemap. IndexNow, a protocol for notifying search engines about new or updated URLs, is another way to speed up discovery.

A caution: ranking isn’t everything. After the Gemini 3 upgrade in January 2026, Ahrefs found only 38% of citations come from top-10 pages. Traditional ranking helps, but it isn’t sufficient alone.

Step 3: Publish things worth citing

AI tools extract facts. Give them facts only you have.

Original data is the strongest lever. One 2026 report that reviewed many studies found original research and proprietary data was the most consistent factor, with citation rates of 38–65% versus 6–15% for standard blog content. Another source says adding original data can raise AI visibility by about 37%. These come from vendors and agencies, so don’t take the exact percentages as gospel, but the direction is consistent across sources.

What counts as original data? You don’t need a research lab:

  • A survey of 100 of your customers
  • Patterns from your own product data (anonymised)
  • A price comparison you actually checked
  • Test results from products you used

Write answer-first. One playbook suggests opening paragraphs of 40–60 words that directly answer the question. A Princeton study cited by another guide says content with clear questions and direct answers was 40% more likely to be cited.

Name real people. One analysis says named expert quotes and statistics paired with sources lift citation rates. Put a real author byline and credentials on your pages.

Keep it fresh. An Ahrefs analysis of about 17 million citations found AI-cited URLs are on average roughly 25% newer than Google-organic ones, with ChatGPT showing the strongest freshness preference. But it isn’t universal. Seer’s recency study found Perplexity is the one engine where strong, older evergreen content still earns citations. So refresh your key pages when you have something new to say. Don’t change dates just to look recent. That’s a mistake I’ll come back to.

Step 4: Build your presence off your own site

This is the part people underestimate, and the part that takes longest.

  • Earned media beats paid. As I covered in my article on editorial and sponsored links, Muck Rack’s analysis of more than 25 million links found earned media made up 84% of AI citations, with paid or advertorial content at just 0.3%.
  • Brand mentions matter. Branded web mentions correlated with AI visibility at about 0.66, versus 0.218 for raw backlink acquisition. Correlation, not proof, but it’s a strong hint.
  • ChatGPT leans on Wikipedia and directories. Wikipedia was roughly 16% of ChatGPT citations in one large study, and for local-business queries, ChatGPT sent 44.7% of citations to directories and review platforms. If you qualify for a Wikipedia page (it has strict notability rules), keep it accurate and well-sourced. Never edit it with a conflict of interest without following the rules. Claim and complete your profiles on the review sites that matter in your category.
  • Reddit matters, especially for Perplexity. Reported shares range widely by study, from around 20% to 46.7%, so don’t hold onto a single number. The approach is simple: answer questions honestly in relevant communities, disclose who you are, and don’t spam. Real participation works. Link-dropping gets you banned.
  • Gemini is different. Yext found 52.15% of Gemini chat citations came from brand-owned websites, the highest share of any major engine. So your own site’s About, FAQ and product pages matter more there.
  • Patience. One practitioner estimates awareness efforts like podcasts and talks show up in AI citations on a 6–12 month delay. That’s an anecdote, but it fits how slowly reputation builds.

Step 5: Make it clear who you are

AI tools try to connect names to entities. Confusion costs you.

  • Use the same company name, description and key facts on your site, your social profiles and your directory listings.
  • Add Organization schema and link your official profiles with “sameAs.”
  • Keep author bios consistent.

Now the honest part about schema. Google says no special AI schema or markup is needed. And the research is mixed: one BrightonSEO presentation found schema drove a large increase in Google AI Overview citations but a measurable decrease on ChatGPT, Gemini and Copilot, with no effect on Perplexity. That’s one case, but it shows why I’d use schema for normal reasons (accuracy, rich results) and not treat it as a magic switch.

Step 6: Skip the llms.txt hype

You’ve probably seen “add an llms.txt file” advice. The evidence isn’t there. SE Ranking’s analysis of roughly 300,000 domains found no correlation between having the file and being cited, and there’s no public confirmation from OpenAI, Anthropic or Perplexity that their retrieval systems use it. Google has said it doesn’t use it either. It’s cheap to add, but I wouldn’t spend an afternoon on it.

Step 7: Measure it, simply

You can’t improve what you don’t track.

Manual method (free):

  1. Write 5–10 questions your customers would actually ask.
  2. Once a week, run them in ChatGPT, Perplexity and Gemini (or Google AI Mode).
  3. Log whether you’re mentioned, cited, or absent, and who is cited instead.
  4. Note the date. Answers change.

One practitioner uses exactly this approach with five pillar queries a week, and another tracked 13,184 citations across four LLMs weekly with 12–40 prompts per client. The routine is the same.

Analytics: OAI-SearchBot citations include a utm_source=chatgpt.com parameter, so you can see ChatGPT referral traffic in your analytics. Google also introduced a generative AI performance report in Search Console in June 2026.

Tools: Ahrefs and Semrush both shipped generative engine tracking this year, and Profound is a specialised tracker. Start with the manual method, and pay only if you need scale.

Timeline: Some vendors say Perplexity reflects changes within about 6 weeks, while ChatGPT and Google tend to move slower. Treat these as rough.

Mistakes to avoid

Blocking the wrong bot. Check robots.txt first.

Faking freshness. Changing “Updated October 2026” without actually updating the page is easy to do and easy to spot. Update the content, then the date.

Buying sponsored posts and expecting citations. The 0.3% figure is the reminder.

Spamming Reddit and Quora. It backfires and can hurt your reputation.

Testing once. AI answers vary run to run. Test weekly over a month before concluding anything.

Chasing one platform. With 11% domain overlap, a strategy for ChatGPT alone leaves Perplexity and Gemini untouched.

Ignoring what AI says about you. If an answer gets your pricing or features wrong, find the source page (an old directory listing, an outdated review) and fix that.

Believing every statistic. Many of the percentages floating around come from companies selling AI visibility tools. Use them as direction, not promises.

A simple 30-day plan

  • Week 1: Check robots.txt, set up Bing Webmaster Tools, confirm pages are indexed.
  • Week 2: Run your baseline prompts and log results. Fix inconsistent brand info across profiles.
  • Week 3: Publish one piece of original data or a thoroughly updated key page, answer-first, with a named author.
  • Week 4: Share it where people discuss the topic (honestly), pitch it to one relevant writer, and re-run your prompts.

Final thoughts

Nobody can guarantee a citation. The tools are changing, the studies conflict, and anyone who promises a ranking in ChatGPT is selling something. But the pattern keeps repeating: be reachable by the crawlers, say something worth quoting, and be talked about by people other than yourself.

Start with the boring check. Open your robots.txt today and make sure you aren’t blocking the bots you want.

Sources: