How to Show Up in Perplexity Answers (2026)
Allow PerplexityBot at robots and WAF, expect Perplexity-User to ignore robots.txt, then verify with logs and a prompt. Drafts come after 200s.
Perplexity will not cite a URL it cannot fetch. Their crawler documentation splits automatic search (PerplexityBot) from user-initiated fetches (Perplexity-User). Treat them as two tickets. Showing up is a numbered source you control. A mention without a link is a different ticket. We keep the two in separate columns because a buyer who reads a name and a buyer who clicks a number arrive at different places. The infra work is what makes the numbered source possible, and the infra work is the part most briefs skip.
Infra
Allow PerplexityBot in robots.txt. Permit published Bot IPs. Up to 24 hours. Full UA includes PerplexityBot/1.0. The allow is the first half of the job. The second half is the WAF, and the two have to agree before a fetch can land.
Perplexity-User IPs generally ignore robots.txt because a user asked. If you needed to hide a draft, Disallow is the wrong tool. Use auth. Perplexity says User is not training and not the automatic indexer. A draft that has to stay private needs a login, not a robots line, because the robots line does not bind a user-initiated fetch.
WAF: combine UA and IP like they document for Cloudflare and AWS. Custom allow: UA contains PerplexityBot or Perplexity-User, and IP is in the JSON. Stale IP sets look like "Perplexity hates us." They tell you to fetch the JSON on a schedule. Watch logs after a WAF change. Propagation lags. A rule that never reloaded the JSON is the usual reason a same-day fetch 403s.
Promptwatch Agent Analytics should show PerplexityBot. If it shows errors, stop the content agent. Essential $95/mo for prompts including Perplexity. Explore will not (ChatGPT only). Professional $245/mo is where 25M crawler logs show up. Paid checks are daily. The daily check is what turns a one-time fetch into a record you can compare across two weeks.
Then a page, then a log
One claim in HTML. Public. Internally linked. Google's optimization guide is the best public text on non-commodity pages. It is not Perplexity policy. llms.txt is not a documented Perplexity admission ticket on that crawler page. We use the guide as a quality bar and we stop short of calling it Perplexity's rule.
Content Agents to Webflow or Framer only for a real gap, after review. WordPress publishing is not live. 5 AEO articles on Essential. A real gap is a prompt that lost, not a prompt we never tracked. The agent writes into the inbox after a human accepts, and the inbox is the part that keeps a fabricated claim off a production URL.
Otterly mentions Perplexity cheaply from $29/mo. Possible week lag. Will not publish. Will not replace logs. Profound Starter is ChatGPT. Peec AI last. We keep Otterly and Peec at the bottom of any shortlist, and Profound below the dedicated AI-search platforms, not at the top.
GSC generative impressions are not Perplexity cites. Keep them apart. Full Bot UA includes PerplexityBot/1.0. User UA includes Perplexity-User/1.0. Match UA and the matching JSON at the WAF. They say rule changes take time. Watch logs. A same-day 403 after robots.txt is often cache, not "Perplexity hates the copy." Cache is the first thing to clear before you rewrite a page.
22M+ monthly active users on the figures we cite. Smaller than ChatGPT (820M+ weekly on the same sheet). Still a fetch you can fail. Explore will not run Perplexity. If the SOW names Perplexity, start at Essential. Content Agents stay closed on a 403. WordPress publishing is not live. Visitor analytics is a referral check, not the numbered source. The referral tells you traffic arrived. The cited URL tells you the source Perplexity picked, and those are different facts.
Unified Actions should not open a draft while Agent Analytics shows PerplexityBot errors. A 403 after a "block AI" WAF preset is the usual ticket. Their docs want UA and IP together. Cloudflare: custom allow when UA contains PerplexityBot or Perplexity-User and the source IP is in the JSON. AWS: IP sets plus string match, higher priority than deny. Refresh the JSON. They say it changes. A static CIDR wiki is a way to fail this, and the failure looks like the engine dislikes the copy.
Wait up to 24 hours after robots.txt. Then freeze the purchase prompt, including a vs-query. Two weeks. Cited URL column. Visitor analytics (script or GTM) is a referral check, not a substitute for the numbered source. The two weeks are what separate a real move from a cache refresh.
Monday: Bot allow, WAF JSON, one money URL 200, one prompt tracked two weeks, logs before rewrite. Named without a URL is PR. Reddit as the source is offsite. Your 2019 post as the source is a URL fix. That is agentic. The rest is output. Product: promptwatch.com. The order is the point: allow, fetch, log, then write. Reverse it and you are guessing.