B2B buyers now ask ChatGPT and Perplexity the questions they used to ask Google or a sales rep: which tools fit a 50-person team, what a product costs, how long implementation takes, what the alternatives are. Getting cited in those answers takes three things: the assistant's crawler can read your pages, your pages answer those exact questions, and other sources back you up.
Step 1: Let the right crawlers in
Both products keep their own search index:
- ChatGPT search uses OAI-SearchBot. OpenAI says sites that block OAI-SearchBot won't be shown in ChatGPT search answers. ChatGPT-User fetches pages when a user asks ChatGPT to open a link. GPTBot is the training crawler, and blocking it doesn't affect search.
- Perplexity uses PerplexityBot for its index and Perplexity-User for live fetches.
Check three places:
robots.txt: noDisallowrule that catches these bots, including broad rules for all user agents on folders like/resources/or/blog/.- Your CDN and firewall: Cloudflare and other providers have settings that block AI crawlers, and Cloudflare turned blocking on by default for new domains in 2025. Allow the search bots you want.
- Your server logs: confirm the bots get 200 responses on your important pages. See how to analyze server log files.
Also make sure your main content is in the HTML the server sends. Many marketing sites built as single-page apps show an empty shell to crawlers that don't run JavaScript, and most AI crawlers don't.
Step 2: Map the questions buyers ask
List the questions that come up in sales calls, demos and support tickets, then group them by buying stage:
| Stage | Example prompt | Page that should answer it |
|---|---|---|
| Problem | "how do I reduce invoice approval time" | Guide that names the causes and fixes, with your product as one option |
| Category | "best procurement software for mid-size manufacturers in India" | Use-case page for that segment, with clear fit criteria |
| Comparison | "[you] vs [competitor]" | Honest comparison page with a table |
| Evaluation | "how much does [you] cost" / "does [you] integrate with Tally" | Pricing page with real numbers, integration pages |
| Risk | "how long does [you] take to implement" | Implementation page with timeline, steps and requirements |
Assistants run several searches for one prompt, a technique often called query fan-out. A page that covers a sub-question cleanly can be cited even if it doesn't rank for the main query.
Step 3: Write so the answer can be lifted
- Open each section with the answer in one or two sentences. Put background after it.
- Use the buyer's wording in headings, including "vs", "cost", "best for" and "how long".
- Give specifics: price ranges, contract terms, setup time in days, limits on users or volume, the regions you serve. Vague claims ("flexible pricing", "fast setup") give a model nothing to quote.
- Use tables for comparisons and feature lists. Models handle them well.
- State who the product isn't for. Honest fit criteria make comparison pages more credible, and assistants often repeat them.
- Show a date and keep it current. For pricing and feature questions, assistants favour recent pages.
Step 4: Publish something only you have
Most B2B content restates what ranks already, and assistants have no reason to cite a restatement. Original material gets cited:
- Benchmarks from your product data, with the method explained
- A yearly survey of your customers or your industry
- Anonymised before-and-after numbers from implementations
- Calculators and templates with the assumptions written out
One solid data page can earn citations for years, and it gives journalists and other writers a reason to link to you.
Step 5: Get mentioned where assistants look
When a prompt asks for recommendations, assistants lean on third-party sources: review platforms such as G2 and Capterra, Reddit and community forums, industry publications, analyst lists and "best X" articles. Check which sources Perplexity cites for your category prompts (it shows them openly), then:
- Ask happy customers for reviews on the platforms that appear
- Get listed, and listed accurately, in the comparison articles that appear
- Answer questions in the communities that appear, as yourself, without spamming links
Step 6: Measure and repeat
Run a fixed set of 30 to 50 buyer prompts in ChatGPT and Perplexity every month and record whether you're cited, mentioned and described correctly.
The citations themselves are worth analysing. For each prompt, note every URL the answer cites, then group them by domain. The domains that come up again and again for your buyers' questions are the sources the assistants trust in your category: those are the sites to be mentioned on in step 5, and the pages of yours that never appear are the ones to rewrite in step 3. A spreadsheet is enough for a few dozen prompts; paid trackers do the same at scale. In GA4, segment sessions referred by chatgpt.com and perplexity.ai. ChatGPT also tags many links with utm_source=chatgpt.com. Then rewrite the pages that should win a prompt but don't. How to check your AI search visibility has the full method.
If your product has public documentation, it's often the most cited part of a B2B site. I cover that separately in optimizing documentation for ChatGPT and Perplexity.

