How to Get Your Site Indexed and Cited by ChatGPT: a 9-Step Tutorial

In short

To get indexed and cited by ChatGPT, you need four things. Your site must be reachable by OAI-SearchBot and GPTBot: according to Otterly.ai, 73% of sites have technical barriers for AI crawlers, including robots.txt blocks, CDN restrictions or rendering only through JavaScript. Content must be delivered in plain HTML, not rendered only through JavaScript. Pages must be visible to Bing, which is one of the data providers behind ChatGPT Search alongside OpenAI’s own crawler. And the brand must be mentioned on other sites: according to AirOps, roughly 85% of a brand’s visibility in AI answers is formed off its own site.

Page structure decides whether it gets quoted once it’s found: a direct answer in the first 100 words, definitions, numbered steps, tables and a dated update with a named author. Schema markup and llms.txt aren’t worth the effort. Nobody can promise a position: according to AirOps, only about 30% of brands hold their place in two consecutive answers. Here is the nine-step sequence we run.

Step 1. Open robots.txt to the AI crawlers

This is the fastest fix and the most common mistake. Technical barriers for AI crawlers (robots.txt blocks, CDN restrictions or JavaScript rendering) exist on 73% of sites, and the most common cause is robots.txt: a CMS template or a security plugin added a default rule. If OAI-SearchBot cannot fetch the page, nothing else in this guide matters.

Open https://yourdomain.com/robots.txt and make sure it contains these blocks, or at least does not contain a Disallow: / for these agents:

User-agent: OAI-SearchBot
Allow: /

User-agent: GPTBot
Allow: /

User-agent: ClaudeBot
Allow: /

User-agent: PerplexityBot
Allow: /

Which bot does what:

Two more checks after you edit the file:

• Order of rules. A generic User-agent: * block with Disallow: / above these lines does not override them, but a specific Disallow for GPTBot further down the file does. Search the whole file for each bot name.
• CDN and firewall. Some CDNs and bot-protection services block AI crawlers at the network level regardless of robots.txt. Check the bot settings in your CDN and the server logs for the user agents above. If they still haven’t shown up in the logs after a couple of weeks, something upstream is blocking them.

For reference, move.marketing is open to all of these bots, and this is the first item on our own checklist for every client site.

Step 2. Put the content in HTML, not JS-only

Crawlers that feed AI answers read the HTML they receive. If your key text appears only after a JavaScript framework renders it in the browser, there’s a real chance the crawler sees an empty shell.

How to check in two minutes:

1. Open the page in a browser, right-click and choose View page source (the source, not Inspect). In Chrome that’s Ctrl+U on Windows or Cmd+Option+U on Mac.
2. Search the source (Ctrl+F) for a full sentence from the main body of the page, for example the first sentence of your service description.
3. If the sentence is there, the content is in HTML. If it isn’t, and you only see script tags and a root <div>, the content is JS-only.

Command-line alternative: curl -L https://yourdomain.com/page | grep "your sentence".

If the content is JS-only, the fix is server-side rendering or static generation for the pages that matter: the home page, service and product pages, the key articles. Most modern frameworks support this; the work is turning it on and testing it. Do this before you invest in content, or the content will never reach the index.

Step 3. Get into Bing: Webmaster Tools, sitemap, IndexNow

ChatGPT Search uses Bing as one of its data providers, alongside OpenAI’s own OAI-SearchBot crawler, according to OpenAI’s official documentation. A page missing from Bing’s index has a lower chance of showing up in ChatGPT Search results. Many companies have never opened Bing Webmaster Tools, and this is the second cheapest win after robots.txt.

1. Register at Bing Webmaster Tools. You can import verified sites from Google Search Console, which saves the verification step.
2. Submit your XML sitemap. Make sure it lists only canonical, indexable URLs, and that the lastmod dates show the page’s real update time, not the date of the last full site rebuild.
3. Set up IndexNow. Generate a key, place the key file at the root of the domain, and send a ping every time a page is published or updated. Most CMS platforms have an IndexNow plugin or a built-in option; on custom stacks it’s a single HTTP request per URL. IndexNow helps Bing find pages faster; there’s no direct confirmation that this speeds up how a page appears in ChatGPT.
4. Check the coverage report after two weeks. Find pages Bing found but excluded, and fix the reasons (redirect chains, a leftover noindex from staging, duplicate content).

IndexNow speeds up discovery, but it doesn’t force indexing. Pages still have to be worth indexing, which is the next step.

Step 4. Structure pages so they get quoted

Once a page is reachable and indexed, whether it gets quoted depends on structure more than on length. Freshness and clarity (direct answers, tables, definitions, numbers with a source) matter more than word count. What we apply to every page we want to see cited:

• A direct answer in the first 100 words. The first paragraph answers the question in the title with no warm-up. If the page is titled “How much does a dental implant cost in Warsaw,” the first paragraph gives the range, the factors and the date.
• Definitions. When you introduce a term, define it in one sentence in a form that can be lifted out of context: “Share of Voice in AI is the share of answers to a set of prompts in which your brand is mentioned.”
• Numbered steps. Processes as ordered lists, one action per item, with the expected result.
• Tables. Comparisons, prices, timelines and specifications as real HTML tables, not images.
• Numbers with a source and a date. A figure with a note like “according to X, September 2026” is quotable; a figure without a source is not.
• A dated update. A visible “Updated on” date that changes when the content actually changes, with money pages reviewed at least once a quarter.
• A named author with a role. A real person, their role and a short bio. For a clinic that’s the doctor; for a SaaS it’s the product lead.
• One topic per page. A page that answers one question well gets quoted more often than one that touches ten.

Reread your key pages with one question: if an assistant had to answer using only this page, could it lift a clean sentence, a number and a list of steps? If not, restructure it.

Step 5. The off-site 85%: mentions, reviews, YouTube, Wikipedia, Reddit

About 85% of a brand’s visibility in AI answers is formed off its own site: media, directories, reviews, forums and YouTube. Brand mentions on third-party sites correlate with visibility in Google’s AI Overviews roughly three times more strongly than backlinks do, according to Ahrefs. That’s why a technically flawless site with no external footprint stays invisible.

Where to work, in order of increasing effort:

1. Directories and listings. Industry directories, local business listings, comparison sites in your category. The company name, address, description and category should be identical everywhere.
2. Media. Expert comments, guest articles, case studies in trade and regional outlets. One well-placed article with your brand name and a clear description of what you do is worth more than a dozen backlinks from irrelevant sites.
3. Reviews. Google, industry review platforms, app stores. Volume, recency and text that names specific services matter, because answers pull details straight from review text.
4. YouTube. Titles and descriptions that clearly state what the company does and for whom. Videos get indexed and cited.
5. Wikipedia and Wikidata hygiene. Don’t write a promotional article about yourself. Do check that existing mentions of your company, founders and products are accurate, and that a Wikidata item, if one exists, has the correct name, website, founding date and industry.
6. Reddit, honestly. Reddit is the most-cited domain in AI answers, according to Peec AI, which analyzed 30 million sources across ChatGPT, Google AI Mode, Gemini, Perplexity and AI Overviews. Reddit itself has reported removing an average of 100,000 automated accounts a day as part of its anti-bot effort, so the platform actively cleans out fake activity. Only one approach works: a real employee account, a real name and useful answers in relevant threads, with the company mentioned only where it’s directly relevant. No fake questions, no paid posters.

Assign one person to this stream and give them a monthly quota of new mentions. It compounds slowly, then shows up everywhere at once.

Step 6. What not to bother with

Time is limited, so here’s what we deliberately leave out of AI visibility work:

• Schema.org and FAQ markup as a GEO tool. A study by Ahrefs tracked 1,885 pages that added JSON-LD markup and found citation gains statistically indistinguishable from zero: Google AI Overviews even dropped 4.6%, AI Mode grew 2.4%, ChatGPT grew 2.2%. Keep the markup as ordinary SEO hygiene for rich results, but don’t buy it as a ticket into ChatGPT, and don’t let anyone sell it to you as one.
• llms.txt. John Mueller from Google has publicly said Google does not use or support llms.txt, and there’s no evidence it changes how ChatGPT chooses sources. Publishing one does no harm; expecting results from it does.
• Content volume for its own sake. Ten new pages a week with no direct answers, sources or external mentions won’t move Share of Voice.
• Blocking bots to “stop competitors from stealing” or “AI-proof” your content. If you want to be cited, you want to be read.
• Promises of a position. Any offer that promises a specific place in ChatGPT answers is selling something that can’t be delivered. More on this in Step 9.

Step 7. How to check: 20 buying prompts and Share of Voice

You can’t manage what you don’t measure, and there’s no Search Console for ChatGPT. Build your own panel:

1. Write 20 buying prompts. Skip brand queries and write the questions a buyer asks before choosing. For a clinic: “best clinic for laser resurfacing in Warsaw,” “how to choose a dermatologist for acne scars.” For a SaaS: “billing service for a small agency working with EU clients.” Mix the stages: researching, comparing, ready to buy.
2. Run them in ChatGPT and Copilot (and Perplexity if it matters for your market), in a fresh chat each time, at least once a month.
3. Record for each prompt: whether your brand was mentioned, whether it was linked, which competitors were mentioned, and what source was cited.
4. Calculate Share of Voice: the number of prompts where you were mentioned divided by the total. Track it by prompt group and by assistant.
5. Keep a list of sources. The domains that get cited on your prompts are your mention targets from Step 5.

Tools that automate this: Otterly from $29 a month, Peec at €89 to €199 a month, Profound from $99 a month. A spreadsheet with 20 rows and monthly columns is enough to start; move to a tool once you’re seriously tracking more than one market or more than one assistant.

If you also run paid placements in ChatGPT, keep the organic Share of Voice panel separate from ad reporting, or you won’t know what’s actually working. We covered the paid side in how to launch ChatGPT Ads step by step.

Step 8. The 30/60/90 plan

The technical column is mostly closed out by day 30. From there the work moves to content structure and mentions, in that order of speed and the reverse order of impact.

Step 9. Why no one can promise a position

Answers in ChatGPT are generated fresh each time, and the set of sources shifts. Only about 30% of brands hold their place in two consecutive answers to the same prompt. That’s how the medium works, not a sign something’s broken. Here are three consequences for how you buy this work:

• Measure a share, not a position. Share of Voice across 20 prompts, tracked monthly, is a stable number. A “position” in a single answer is not.
• Measure assisted conversions, not traffic. ChatGPT often names a brand without linking to it. Watch branded search, direct traffic and a “how did you hear about us” field alongside referral clicks.
• Fix the metrics in the contract. A serious agreement names the prompt set, the assistants, the reading schedule and the targets for Share of Voice and assisted conversions. It does not promise a place in an answer.

At MOVE, AI visibility work starts with an audit from $1,500: crawler access, HTML rendering, Bing indexation, money page structure, your external mention footprint and a Share of Voice baseline across 20 prompts, followed by a 30/60/90 plan with owners assigned. If you want to know where your site stands before you spend on content or ads, see AI advertising and visibility at MOVE or book an AI visibility audit.