How to get your startup recommended by ChatGPT and AI search (2026)

There's no published formula for getting recommended by ChatGPT, Perplexity, Claude, Copilot or Google's AI Overviews. What the engines do publish is narrower and more useful: which crawlers they use, how to let them in, and (for Google and Bing) a plain statement that ordinary search fundamentals are what get a page cited. So the job is to be crawlable, be in the indexes, say the same facts about yourself everywhere, and publish pages that are worth quoting.

This guide sticks to what the engines document, then shows our own starting point: a 48-prompt test where Remote Builders showed up in zero answers. We don't sell an AEO tool, so you get the unflattering numbers too.

What the engines actually document

Each engine documents its crawlers and opt-outs, but none documents how it ranks or picks citations. Here's what each one says, checked on 26 October 2026.

ChatGPT. OpenAI's crawler overview says OAI-SearchBot "is used to surface websites in search results in ChatGPT's search features", that opted-out sites "will not be shown in ChatGPT search answers, though can still appear as navigational links", and that robots.txt changes take about 24 hours to reach its search systems. The settings are independent: you can allow OAI-SearchBot and disallow GPTBot, which covers training. ChatGPT-User, which fetches pages when a user asks, "is not used to determine whether content may appear in Search". OpenAI's help centre adds that ChatGPT search "sometimes partners with other search providers", rewriting your question into targeted queries, and lists Microsoft's privacy statement among those providers. That help page blocked our fetcher, so we read it through the r.jina.ai text reader.

Claude. Anthropic's crawler article lists three bots. ClaudeBot collects content that could contribute to training. Claude-SearchBot indexes for search, and disabling it "may reduce your site's visibility and accuracy in user search results". Claude-User fetches pages when someone asks Claude a question.

Perplexity. Perplexity's crawler docs say PerplexityBot is "designed to surface and link websites in search results on Perplexity" and isn't used to crawl content for AI foundation models. Perplexity-User fetches pages on a user's request and "generally ignores robots.txt rules". Changes can take up to 24 hours.

Google. The Search Central page on AI features is the most direct statement any engine makes: "There are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary." A page must be indexed and eligible to show with a snippet. Google also says AI Overviews and AI Mode may use "query fan-out", issuing several related searches across subtopics, so a page that answers one sub-question well can be pulled in. Blocking Google-Extended limits Gemini training and grounding in Gemini Apps, and Google says it doesn't affect inclusion or ranking in Google Search.

Copilot and Bing. Bing's Webmaster Guidelines now cover Copilot and "grounding" (the pages an AI answer is built from). They say Bing and Copilot "rely on the same core crawling, indexing, and ranking foundation as traditional search", and that "GEO does not guarantee grounding or citations in AI experiences". We read this page through r.jina.ai as well.

What you control

You control six things: crawl access, index coverage, consistent entity facts, answer-first pages, an optional llms.txt and honest third-party mentions. None guarantees a recommendation. All of them are documented or low-risk.

1. Let the search crawlers in

Check your robots.txt allows the search and answer crawlers named above, and that your CDN or firewall isn't blocking them. Google lists "ensuring that crawling is allowed in robots.txt, and by any CDN or hosting infrastructure" as a best practice for AI features. A minimal set looks like this:

• User-agent: OAI-SearchBot / Allow: /
• User-agent: Claude-SearchBot / Allow: /
• User-agent: PerplexityBot / Allow: /
• Googlebot and Bing's crawler left unblocked

Decide training crawlers (GPTBot, ClaudeBot, Google-Extended) separately. Blocking them is a training choice, and none of the vendors says it affects search inclusion. Anthropic notes that blocking by IP address may not work reliably, because it stops the bot reading your robots.txt.

2. Get into Bing's and Google's indexes

Verify your site in Google Search Console and Bing Webmaster Tools, submit a sitemap to both, and set up IndexNow. Bing's guidelines list IndexNow and XML sitemaps as discovery signals and say to "Notify IndexNow when content is added, updated, or deleted." Google, by contrast, says recrawling can take "anywhere from several days to several months".

IndexNow works like this: host a text file named after your key at your site root, then send changed URLs to a participating engine, up to 10,000 per POST. Submitted URLs are "automatically shared with all other participating search engines". The participant list on 26 October 2026 was Bing, Yandex, Seznam, Naver, Yep, Internet Archive and Amazonbot. Google isn't on it.

Bing ties this to Copilot directly: duplicate URLs "reduce Bing's confidence in selecting a URL for grounding results or citations", so use canonical tags that point each page to itself.

3. Make your entity facts match everywhere

Use one name, one description and the same key facts on your site, your profiles and directory listings. Bing's guidelines say "Clear entity definition improves grounding visibility and citation accuracy" and ask for "clear and consistent naming for people, organizations, products, and locations". Our own test (below) found this was a real problem for us: name variants and conflicting member counts across the website, llms.txt and Meetup.

If your name collides with something else, add a disambiguator every time. "Remote Builders" collides with developer-tool documentation, so we write "Remote Builders community and podcast".

4. Write pages that answer one question, with dated sources

Give each page one question, answer it in the first lines, and back each claim with a dated source. That's close to word for word what Bing asks for: "Focus Each URL on a Single Topic", "Surface Key information early", make facts and definitions explicit, and keep content accurate and up to date. Bing's AI Performance help page adds "Support claims with evidence" and suggests "descriptive headings, concise sections, tables, and FAQ-style content".

Google's version is shorter: follow the SEO fundamentals, keep important content in text, and make sure structured data matches the visible page. It also says "There's also no special schema.org structured data that you need to add."

5. llms.txt: optional and unproven

Add an llms.txt if you like, but don't expect it to move anything. llms.txt is "A proposal to standardise on using an /llms.txt file to provide information to help agents use a website", published by Jeremy Howard on 3 September 2024. On 26 October 2026 we found nothing in OpenAI's, Anthropic's, Perplexity's, Google's or Bing's documentation saying their search uses it. Google says "You don't need to create new machine readable files, AI text files, or markup to appear in these features."

Our own data point: remotebuilders.io had an llms.txt at baseline and still wasn't retrievable by name. That doesn't prove llms.txt is useless. It does show it isn't a substitute for being indexed.

6. Earn third-party mentions and honest reviews

Get listed and talked about on sites the engines already retrieve: directories, community lists, podcasts, comparison pages and review platforms. In our test, every appearance of Bali Startup & Tech came through Meetup, not its own site. Bing warns that "link buying, link spamming, private networks, and artificial social promotion schemes" reduce visibility. Ask real customers for reviews, and ask for honest ones. The first customers guide covers where those early conversations come from. Which listings actually work is the kind of thing Remote Builders members will compare notes on in the members' chat, which is being set up now.

Our own experiment: 48 prompts, one honest starting point

We ran a fixed panel of 48 prompts about our two communities and found almost nothing. Bali Startup & Tech, the Bali community whose Telegram will host the Remote Builders members' chat, appeared in 5 of 48 answers. Remote Builders, an invite-based community for founders who build remotely, appeared in 0 of 48.

Method

The panel is 48 fixed prompts across community, events, AI building, living, visas, coworking, remote business, podcasts and brand questions. The searches ran on the night of 25 October 2026 (Bali time) and the baseline is dated 26 October 2026. Our file states the limits plainly, so we'll repeat them:

• No engine was queried directly. None of ChatGPT, Perplexity, Claude.ai, Gemini or Google AI Overviews was asked. Each prompt was run once through a web-search tool in standard mode (a US-region web index whose provider isn't disclosed), with extended mode used three times for thin results.
• The proxy is rough. We treated the tool's short synthesised answer as a stand-in for a retrieval-based engine. Our notes say that's closest to Claude's web search and only a rough guide to ChatGPT, Perplexity and Google AI Overviews.
• Scoring. For each prompt we recorded the first five relevant URLs, whether each brand appeared, through which URL, and who appeared instead. We then checked the URLs that mattered with a direct fetch.
• Known biases. One run per prompt, a logged-out US-centred index, and "top 5" means the tool's order, not Google's.

What we found

• All five appearances came through meetup.com. Only one was a current, working citation. Others pointed to a Meetup event URL that returned 404 when we checked it.
• Crawl access wasn't the problem. The Bali Startup & Tech robots.txt already allowed GPTBot, OAI-SearchBot, ClaudeBot, PerplexityBot and Google-Extended.
• Our own pages weren't the ones being cited. The audit found site-level fixes on our side, mainly canonical tags and facts (names and member counts) that didn't match across our site and profiles. Those fixes are in the checklist below.
• Remote Builders wasn't retrievable at all, even when the prompt named remotebuilders.io. Results went to developer-tool documentation and to SafetyWing's "Building Remotely" podcast. The site was reachable, allowed all crawlers, and had a sitemap, an llms.txt and a self-referencing canonical. It just wasn't in the index the tool used.

What happens next

We rerun the panel on the first Monday of each month, and we'll update this page with what moved and what didn't. The fixes are the boring ones from the list above: self-referencing canonicals, sitemap and IndexNow submission, one name and one set of facts everywhere, and free directory listings. We'll also add a manual check: 20 of the prompts asked in fresh, logged-out or temporary sessions on ChatGPT, Perplexity, Claude and Google, which is the real engine baseline. We won't claim a result until we have one.

What doesn't work, or carries risk

Unsourced statistics, fake reviews and manipulation tricks are the three things to avoid. The first wastes your time, and the other two can cost you money or visibility.

Unsourced AEO statistics. Many AEO vendor posts quote dramatic numbers without a dataset. We couldn't verify most of the common figures, so they're not on this page. The one we could trace is the "23x" conversion claim. It comes from an Ahrefs post by Patrick Stox (16 June 2025) reporting Ahrefs' own site over 30 days: "AI search traffic accounts for just 0.5% of our traffic", and "12.1% of signups" came from it. That's real data from one SEO tool company's site. It isn't a benchmark for yours.

Buying fake reviews. In the US, the FTC's final rule (announced 14 August 2024) bans fake reviews and testimonials, including AI-generated ones and reviews from people with no real experience of the business. It also bans incentives conditioned on positive or negative sentiment, undisclosed insider reviews, company-controlled "independent" review sites and buying fake followers or views. In December 2025 the FTC warned 10 companies that violations can bring "civil penalties of up to $53,088 per violation" (US dollars).

In Australia, the ACCC's online reviews guidance says it's against the law to create fake or misleading reviews or arrange for others to. Incentives must apply whether the review is positive or negative and must be disclosed. The Federal Court ordered HealthEngine to pay A$2.9 million in penalties, partly for publishing misleading patient reviews.

Prompt injection and keyword stuffing. Bing lists "Prompt Injection and AI Manipulation" and "Artificially Engineered Language" written to "trigger citations or AI responses" as practices that can reduce visibility or get you removed. Hidden instructions to AI models are a bad bet.

How to measure it

Measure with a fixed prompt panel run monthly, referral traffic from the AI engines, and signups by source. Visibility only matters if it turns into signups.

• A fixed prompt panel. Write 20 to 50 questions your buyers actually ask, freeze the wording, and run them on the same day each month. Record the date, engine, location and whether you were logged in. Score each answer: 0 not mentioned, 1 mentioned, 2 linked to a URL you control, 3 recommended first. Note the URL cited and any wrong facts. Add new prompts with new IDs rather than editing old ones, so the trend stays comparable. ChatGPT's help centre says people who aren't signed in can also use web search, which makes clean runs easier.
• Referral traffic. Filter your analytics referrers for chatgpt.com and perplexity.ai. Under the default strict-origin-when-cross-origin policy, MDN says browsers send only the origin cross-site, and a no-referrer policy sends nothing. So you'll see which engine sent the visit, never the question, and some visits will show as direct. Treat the count as a floor.
• Search Console and Bing. Google counts AI Overviews and AI Mode inside the normal Performance report under the "Web" search type, so you can't separate them there. Bing Webmaster Tools has an AI Performance report showing citations and "grounding queries" for supported AI experiences. Bing says citations don't represent clicks or traffic, and the data is sampled.
• Signups by source. Ask "how did you hear about us?" on your signup form and keep the free-text answers. People often type "ChatGPT" there when the referrer is missing.

Common mistakes

• Blocking OAI-SearchBot while trying to block training. GPTBot is the training crawler. Blocking the wrong one removes you from ChatGPT search answers.
• A site-wide canonical pointing at the homepage. It quietly tells search engines your other pages don't count.
• Skipping Bing. ChatGPT search draws on third-party search providers including Microsoft, and Copilot runs on Bing's index.
• Inconsistent names and numbers. Three member counts and four spellings make you harder to recognise as one entity.
• Treating llms.txt as the fix. No engine documents using it for search.
• Quoting vendor stats as fact. Trace every number to its dataset or leave it out.
• Buying reviews or offering rewards for five stars. Both are illegal under the FTC rule and risky under ACCC guidance.
• Measuring visibility without signups. A mention that brings no customers isn't the goal.

Checklist: getting your startup recommended by AI search

1. Allow OAI-SearchBot, Claude-SearchBot, PerplexityBot, Googlebot and Bing's crawler in robots.txt, and check your CDN or firewall isn't blocking them.
2. Decide GPTBot, ClaudeBot and Google-Extended separately as a training choice.
3. Verify your site in Google Search Console and Bing Webmaster Tools and submit a sitemap to both.
4. Set up IndexNow and ping it whenever a page is added, updated or deleted.
5. Make every page's canonical tag point to itself.
6. Pick one name and one description, and use them everywhere, with a disambiguator if your name collides with something else.
7. Put the same dated key facts on your site, profiles and directory listings.
8. Give each page one question and answer it in the first lines, with dated sources.
9. Add an llms.txt if you like, without expecting it to change anything.
10. Get listed on free directories and community lists, and ask real customers for honest reviews.
11. Freeze a prompt panel and run it on the same day each month.
12. Track chatgpt.com and perplexity.ai referrals, and signups by source, every month.