Ask ChatGPT for "the best tools for turning text into songs" or "an influencer agency that works with AI startups" and you'll get a short list. Some brands show up again and again. Others with perfectly good products never appear.
There's no form to fill in and no switch to flip. What you can do is remove the reasons an assistant would skip you, and give it good reasons to include you. Here's how we approach it.
Why does ChatGPT recommend some brands and not others?
When ChatGPT or Perplexity answers a question about products or vendors, it usually searches the web, reads a set of pages and writes an answer based on them. It can also draw on what the underlying model learned in training.
That means three things decide whether you get named:
- Whether the assistant's crawler could fetch your pages at all.
- Whether the pages it fetched clearly say what you do, who it's for and why it's good.
- Whether other sources it reads agree with that description.
Most brands that are missing from AI answers fail on at least one of these. The fixes are different for each, so it helps to check them in order.
Can ChatGPT and Perplexity reach your site?
Start with robots.txt. Each AI company publishes the names of its crawlers. For ChatGPT, the one that matters most for recommendations is OAI-SearchBot. OpenAI's documentation says sites that are opted out of OAI-SearchBot won't be shown in ChatGPT search answers, and recommends allowing it (OpenAI). GPTBot is a separate crawler used for model training, and you can allow or block it independently.
For Perplexity, the crawler is PerplexityBot, which Perplexity says it uses to surface and link websites in search results (Perplexity). Anthropic, Google and Microsoft have their own agents too. A simple rule of thumb: if you want to be recommended, allow the search and user-triggered agents for every assistant your buyers use.
Then check the layer above robots.txt. CDNs and firewalls increasingly ship one-click settings to block AI bots. These can override a perfectly good robots.txt. If you use Cloudflare or a similar service, look at the bot settings and any WAF rules. Perplexity's docs even include instructions for allowing its bots through Cloudflare and AWS firewalls.
Last, look for walls. Pages behind a login, a cookie consent screen that hides content, or an aggressive bot challenge will all look empty to a crawler.
Can they read your pages without JavaScript?
This is the step most modern marketing sites fail without knowing it. Many sites are built as single-page apps. The server sends a nearly empty HTML file and JavaScript fills in the content in the browser.
Google renders JavaScript, so these sites often look fine in Google Search. Many AI crawlers don't. Vercel and MERJ studied AI crawler traffic in 2024 and found that the major AI crawlers they tracked, including OpenAI's and Perplexity's, didn't execute JavaScript (Vercel).
You can test this yourself in a terminal:
curl -s -A "OAI-SearchBot" https://www.yoursite.com/ | grep -i "<h1"
If that returns nothing, the crawler is probably seeing a blank page. The fix is server-side rendering or prerendering, so every route ships as finished HTML with its title, description and main copy already in it.
Does your site answer the questions buyers ask?
Once an assistant can read your pages, it still has to find a passage worth using. That depends on whether you wrote about the thing the buyer asked.
Which prompts should you plan for?
Write down the questions your buyers type into an assistant. For most brands they fall into a few patterns:
- Category questions: "best AI music generator," "top influencer agencies for tech brands"
- Problem questions: "how do I get developers to try my tool"
- Comparison questions: "Brand A vs Brand B," "alternatives to Brand A"
- Trust questions: "is Brand A legit," "Brand A pricing"
Pick 20 to 50 prompts that matter to your pipeline. Those become your test set.
How should a page be written so an AI can quote it?
Put the answer first. If a page is about what your product does, the first two sentences should say what it does, for whom, in plain words. Assistants often lift a short passage, so make sure the passage stands on its own.
Use the same words your buyers use. If people call your category "AI code review," don't only call it "intelligent pull request analysis."
Be specific and consistent. Use your brand name the same way everywhere. State facts you can back up, and link to sources when you make claims. The GEO research paper from Princeton and IIT Delhi found that citing sources, adding quotations and adding statistics all improved how visible a page was in generated answers.
Give each page one job and a clear structure. A single H1, descriptive subheadings (questions work well) and a short FAQ at the end make it easy for a model to find the relevant section. Structured data such as Organization, Service and FAQPage helps, as long as it matches what's visible on the page.
Who else is talking about you?
This is where many technical GEO projects stall. You can fix your own site completely and still be left out, because assistants rarely rely on one source. They read reviews, listicles, forums, news, YouTube and social posts, and they weigh what those sources say.
If nobody else describes you as a leading option in your category, the assistant has little reason to either. The work here looks like classic PR and community building:
- Get included in honest comparison and "best of" articles in your category.
- Encourage real customers to leave reviews on the sites your buyers check.
- Show up where your category is discussed, such as Reddit, Discord communities, YouTube and LinkedIn.
- Work with creators who already have your buyers' attention, so there's a steady flow of people describing what your product does in their own words.
That last one is why Dime combines GEO with creator marketing. We've written more about the mechanics in how creator content feeds AI search visibility.
How do you know if it's working?
Run your prompt set across ChatGPT, Perplexity, Google AI Overviews and Copilot before you change anything. Record whether you're mentioned, where you appear in the list, how you're described and which sources are cited.
Then repeat the same prompts on a schedule, monthly at minimum. AI answers vary from run to run, so look at trends across many prompts. A single answer proves very little. The useful metrics are:
- Mention rate: the share of prompts where you appear
- Share of voice: your mentions compared with competitors
- Citations: which of your pages, and which third-party pages, get linked
- Sentiment and accuracy: whether the description is right and favorable
This is the core of the tracking work we do for clients on our GEO program. If you'd like a baseline for your brand, you can book a call or send us a note.