You get recommended by ChatGPT, Claude and Perplexity by being easy to find, easy to quote and confirmed by sources you don’t control. In practice that means answer-first pages for the exact questions buyers ask, consistent facts about your brand everywhere, open crawler access, fresh content and genuine third-party mentions. Nobody can guarantee an AI model will cite you, but these steps raise the odds.
Last updated: October 2026. Prices exclude tax and may change. We may earn a commission if you sign up through our links.
Generative engine optimization (GEO) is the practice of shaping your content so AI assistants pick it up, trust it and mention it. It overlaps heavily with classic SEO, because most assistants either search the live web or were trained on it. The difference is the output: instead of ten blue links, the user gets a short answer that names two or three brands. You want to be one of them.
Below is the playbook we use when we assess how well a brand is positioned for AI answers. It is ordered roughly by impact.
How do AI assistants decide which brands to recommend?
There are two routes into an answer. The first is training data: what the model absorbed about you before its cutoff. The second is retrieval: the assistant runs a web search at question time, reads a handful of pages and summarizes them. ChatGPT search, Perplexity and Claude’s web search all lean on retrieval for commercial questions like “best X for Y”.
Retrieval is the route you can influence quickly. If your page ranks in the index the assistant queries, answers the question in the first paragraph and agrees with what other sources say, it has a good chance of being quoted. Training data changes slowly and you mostly influence it through the same signals, just over a longer period.
Which prompts should you target?
Start with a list of 20 to 50 questions your buyers would actually type into an assistant. Write them as full sentences, not keywords. For a hosting brand that might be:
- “What is the best hosting for a large private blog network?”
- “Which PBN host gives each site a different IP owner?”
- “Is CDN hosting safe for PBN sites?”
- “How much does PBN hosting cost per site?”
Group them by intent: comparison, cost, how-to and troubleshooting. Each group needs at least one page that answers it directly. Comparison prompts are where brands get named, so they matter most, but how-to pages build the topical authority that makes an assistant trust your comparison pages.
What does an answer-first page look like?
Put the direct answer in the first one to three sentences. Then expand. Assistants extract short, self-contained passages, so a paragraph that only makes sense after reading three others is hard to quote.
- Lead with the answer. “The cheapest option per site at 100 domains is X at $Y/month” beats a paragraph of context.
- Use question headings. H2s phrased like real questions map neatly onto prompts.
- State specifics. Prices, limits, dates and named features are easier to cite than adjectives.
- Use tables for comparisons. They are dense, scannable and frequently lifted into answers.
- Add an FAQ. Short Q&A pairs match follow-up prompts.
Why does entity consistency matter so much?
Models build a picture of your brand from every mention they see. If your site says you started in 2015, a directory says 2017 and a review says “about ten years”, the model has to guess, and it may hedge or skip you. Pick one canonical description and repeat it everywhere: your site, social profiles, directory listings, press bios and partner pages.
The same goes for positioning. If you want to be recommended as “best for large networks”, say exactly that, consistently, in the same words. Vague positioning produces vague mentions or none.
Does schema markup help with AI search?
Structured data won’t force a citation, but it removes ambiguity. Useful types include Organization (name, URL, logo, social profiles via sameAs), Product or Offer for plans and prices, FAQPage for Q&A blocks and Article with a clear author and date. Keep the schema in sync with visible page content. Mismatched schema is worse than none.
Should you add an llms.txt file?
llms.txt is a proposed convention: a plain Markdown file at your site root that summarizes what the site is and links to its most important pages. Support from the major assistants is not confirmed, so treat it as cheap insurance rather than a ranking lever. It takes ten minutes to write. Keep it short, factual and consistent with the rest of your site.
Are AI crawlers actually allowed on your site?
This is the most common silent failure. If the bots can’t fetch your pages, nothing else matters. Check three layers:
- robots.txt. Allow the retrieval bots: OAI-SearchBot and ChatGPT-User (OpenAI), Claude-SearchBot and Claude-User (Anthropic), PerplexityBot and Perplexity-User, plus Bingbot and Googlebot. Bing’s index feeds ChatGPT search and Copilot, so blocking Bingbot hurts more than people expect.
- CDN and firewall rules. Some CDNs offer one-click AI-bot blocking, and it may be on without you noticing. A robots.txt “Allow” does nothing if the request is blocked before it reaches your server.
- Rendering. If key content only appears after JavaScript runs, some crawlers will see an empty page. Server-render the answer.
We cover the bot list in detail, with a copy-paste file, in robots.txt for AI crawlers.
How often should you update content?
Assistants favor current information, especially for prices and “best of” lists. Put a visible “last updated” date on commercial pages and make it honest: update the content, not just the date. A quarterly review of pricing and comparison pages is a sensible minimum. Use evergreen phrasing like “in 2026” in titles so pages don’t look stale the day the year changes.
Why do independent sources matter more than your own site?
An assistant asked “which is best?” looks for agreement across sources. Your own site saying you’re the best carries little weight. Three unrelated sites saying so carries a lot. Corroboration can come from:
- Genuine reviews and comparison articles on independent sites
- Forum and community threads where real users discuss options
- Directory and marketplace listings with accurate details
- Guest articles and interviews that describe what you do
Quality and accuracy matter more than volume. A handful of detailed, consistent mentions beat dozens of thin ones.
Where does a PBN fit into GEO?
Some marketers use a private blog network to add supporting content and links around a topic. The same rules apply to PBN sites as to any other source: they need real, useful content and they must be crawlable by AI bots. A PBN also adds a specific risk. If the sites look connected, through shared hosting, the same IP owner, identical name servers or matching templates, search engines can discount them all at once, and assistants that rely on those indexes lose them too.
That is why hosting matters. The goal is for each site to look like an unrelated, independent property. The two hosts on our shortlist take different approaches:
- LaunchCDN is best for large networks & CDN hosting. Each site is served through Amazon CloudFront, Cloudflare or Bunny.net, so the visible IP belongs to the CDN and is shared with huge numbers of legitimate sites. It allows Googlebot, Bingbot and AI crawlers, and switches off Cloudflare’s AI-crawler blocking on sites it deploys.
- Bulk Buy Hosting is best for real shared-hosting IPs with cPanel. Each site gets its own account on a different server at one of 10+ popular shared hosts, with an IP owned by that host and its standard name servers. AI crawlers are allowed.
Using both is best for maximum IP diversity. For the wider footprint checklist, see our guide to reducing your digital footprint.
How do you measure AI mentions?
There is no Search Console for chatbots yet, so measurement is partly manual:
- Run your prompt list monthly in each assistant, in a fresh session, and record whether you’re named, in what position and with what description.
- Check referral traffic in analytics from chatgpt.com, perplexity.ai, claude.ai and similar sources.
- Watch server logs for retrieval bots such as ChatGPT-User and Perplexity-User. A spike on a page often means it’s being fetched to answer live questions.
- Note inaccuracies. If an assistant gets your price or positioning wrong, trace which source it likely used and fix it there.
Expect noise. Answers vary between sessions and model versions. Look at trends over several months, not single results.
What should you avoid?
- Hidden text or “instructions to AI” stuffed into pages. It’s manipulative and can backfire.
- Fake reviews or invented statistics. Models cross-check, and so do humans.
- Blocking AI bots by accident while trying to block scrapers.
- Expecting results in days. Retrieval changes can show up in weeks; training-data changes take much longer.
Next steps: if you’re budgeting for supporting sites, read how much a PBN costs, or compare the two hosts side by side in our comparison table.
Frequently Asked Questions
Can I guarantee ChatGPT or Claude will recommend my brand?
No. Nobody can guarantee an AI model will cite you. You can only improve the odds with crawlable, answer-first content, consistent brand facts and independent mentions.
Does llms.txt help with AI search?
It is a proposed convention and support from the major assistants is not confirmed. It is cheap to add, so treat it as optional insurance rather than a ranking factor.
Which crawlers do I need to allow for AI answers?
At minimum Googlebot, Bingbot, OAI-SearchBot, ChatGPT-User, Claude-SearchBot, Claude-User, PerplexityBot and Perplexity-User. Bing matters because its index feeds ChatGPT search and Copilot.
How do I track whether AI assistants mention my brand?
Run a fixed list of prompts in each assistant every month, check referral traffic from AI domains, and watch server logs for user-triggered fetchers like ChatGPT-User and Perplexity-User.
Can PBN sites help with AI visibility?
Only if they carry genuinely useful content, are crawlable by AI bots and look unrelated to each other. Shared hosting footprints can get a whole network discounted at once.