← All guides · Published November 14, 2026 · Reviewed November 14, 2026 · By the Ingenious AI team · 7 min read

Should You Block or Allow PerplexityBot?

The short answer: for most content and business sites: allow it. PerplexityBot is a retrieval crawler — a verified entry in AgentBlock's database — whose real-time search grounding is what puts cited sources into Perplexity's answers. It is not a training crawler. Blocking it is coherent only if you object to being summarised even with attribution; the cost is disappearing from an answer engine your customers increasingly consult. Decide with your own referral numbers, not with a blanket rule.

Build my block kit — free · Test a policy in the simulator first

What is PerplexityBot?

Perplexity's retrieval crawler, a verified entry in AgentBlock's database. It performs real-time search grounding for Perplexity's answer engine, so the pages it fetches are candidates for citation in answers — the opposite economic role from a training crawler.

Retrieval is a distribution channel

The mental shift this guide asks for: retrieval crawlers are not taking something from your site; they are the supply chain of a distribution channel. Perplexity's answers cite sources inline, and the cited page gets the click-through, the brand impression and the trust transfer. PerplexityBot's fetches are how your pages enter that pool. Block the bot and you have not protected the content — you have resigned from the channel.

Perplexity leans harder on live retrieval than most answer engines, which makes its bot's fetches unusually well-correlated with real user questions. In site logs, PerplexityBot visits cluster on exactly the pages people ask about — pricing explainers, comparison guides, documentation — a pattern nothing like a training crawler's systematic sweep.

The honest case for blocking

There is one: substitution. If your business sells the answer itself — proprietary research, paid analysis, a data product — then a faithful summary with a citation can satisfy the reader without a visit, and the citation's value does not offset the lost sale. Publishers in that position block retrieval agents across the board, PerplexityBot included, and accept the invisibility as the price of the paywall's integrity.

Notice what the honest case requires: a product whose value is the content, and evidence that summaries substitute rather than refer. Most small business sites fail the second test in their own analytics — answer-engine referrals arrive, convert at ordinary rates, and grow. Run the numbers on your site before borrowing a publisher's posture.

Measuring your citation value first

AgentBlock's impact preview exists for this decision. Feed it a slice of your server log and it reports, per agent, the actual visit counts the draft policy would affect. Pair that with your analytics referral data for Perplexity over the last quarter, and the trade becomes arithmetic: this many fetches, this many referred sessions, this conversion rate — against a training objection that, for a retrieval agent, does not apply.

Do the same exercise per vendor rather than once for the class. Perplexity, OpenAI retrieval, Claude retrieval and Mistral's user fetcher serve different audiences with different referral volumes on different sites. The builder's per-agent flips are granular because the economics are granular.

Allowing is not surrendering control

An allowed PerplexityBot still operates inside your policy's other controls. Path rules can refuse it your checkout, account and admin paths while your public content stays fetchable. Rate caps can bound its volume per minute and per day. The canary path can tell you if something claiming its name probes places it should not. Allow is a posture toward a named, honest agent — not an open door to everything it might ever do.

And the decision is reversible in minutes: flip the agent in the builder, regenerate the kit, verify the new hash. Owners sometimes treat crawler policy as constitutional; it is closer to a pricing page — important, evidence-based, and expected to change as the numbers move.

What PerplexityBot is not

It is not your SEO. PerplexityBot has no role in Google or Bing rankings, and blocking or allowing it moves nothing there. It is also not a training crawler in AgentBlock's classification — its database note reads 'real-time search grounding; drives citations', which is the vendor's own description of the job. If your policy document lists PerplexityBot under 'training blocks', that line is wrong and worth correcting before anyone audits it.

Finally, allowing PerplexityBot does not allow Perplexity's or anyone else's training crawlers — those are separate entries with separate decisions. The retrieval/training split, held consistently, is the difference between a policy and a mood.

Five actions to take today

  1. Classify before deciding. PerplexityBot is retrieval, not training — do not let a training objection make this decision by accident.
  2. Measure your own trade. Run the impact preview on your log and check Perplexity referrals in analytics for the last quarter.
  3. Apply the substitution test. Block only if summaries demonstrably replace sales of the content itself on your site.
  4. Allow with guardrails. Keep path rules on checkout/account paths and sane rate caps in force for allowed retrieval agents.
  5. Review quarterly. Answer-engine referral volumes move fast; a decision that was right at one volume may be wrong at ten times it.

Worked example: a comparison site prices the citation trade

Switchyard Reviews, a fictional software comparison site, considered blocking all retrieval agents during a policy tidy-up. Its numbers argued otherwise: the impact preview showed 3,380 PerplexityBot fetches in 30 days concentrated on 41 comparison pages, and analytics credited Perplexity with 874 referred sessions that month, converting to affiliate clicks at 6.1% — the site's second-best channel rate.

Switchyard allowed PerplexityBot and the other retrieval agents, blocked the training class completely, and put path rules on its admin and preview paths. Perplexity referrals grew to 1,240 sessions in the following quarter as more of its comparisons were indexed, while training refusals logged 22,000 across the blocked nine.

The site's policy note frames the decision in one sentence its owner reviews quarterly: retrieval is distribution we are paid for; training is inventory we are not.

Frequently asked questions

Is PerplexityBot a training crawler?

No. AgentBlock's database classifies PerplexityBot as retrieval, noting it performs real-time search grounding and drives citations. Training crawlers are a separate class of nine entries, and conflating the two classes is the most common crawler-policy error.

If I allow PerplexityBot, can Perplexity train on my site?

PerplexityBot's documented job is retrieval for answers. Training decisions involve the training-class entries. Allowing one retrieval agent does not change any training agent's decision, which is set independently in your policy.

How do I know if Perplexity sends me real traffic?

Check your analytics for Perplexity referrals and compare the trend with PerplexityBot fetches in your log or the impact preview. On content sites the two usually move together; the citation is the mechanism that produces the visit.

Can I allow PerplexityBot on articles but not on my app pages?

Yes — that is what path rules are for. A policy can allow an agent generally while refusing named path patterns such as checkout, account, admin or preview areas, with per-agent and wildcard rules in the Worker's path matrix.

What if Perplexity's answers quote me inaccurately?

That is a content-correction issue with the vendor, not something crawler policy solves selectively — a block removes accurate and inaccurate citations alike. Most sites handle accuracy by keeping the fetched pages themselves current, since live fetches read today's version.

Build my block kit — free · Browse the agent database

Related guides

HONEST SCOPE: this policy blocks KNOWN agent user-agents and behaviors listed below. Sophisticated spoofing can bypass any user-agent block (e.g. an agent sending a stock browser UA, or rotating residential IPs). This is a control, not a guarantee.
← Back