Perplexity filters SEO filler pages out of its index with Decider

Perplexity Decider judges pages as Perplexity crawls them, so brands with SEO filler or a bot wall in front of their pages lose those pages from its index.

What changed

Perplexity Decider, a classification model Perplexity built, judges the quality of web pages while Perplexity crawls them and filters low-quality pages out of the Perplexity index, Perplexity Research wrote on 12 November 2026 in the post releasing Decider v1.3 and Decider Flash v1.3.

The filter removes two groups of pages. The first is pages that did not load correctly for the crawler: error pages, anti-bot walls and CAPTCHAs. The second is pages that load and hold no helpful information, such as spam, SEO filler and navigation-only pages. “These low-quality pages need to be identified and filtered out of our index,” Perplexity Research, the company’s research team, wrote.

In the same pass over the page, Decider answers other questions about each page: its domain, its intended audience and whether it contains harmful, illegal or personally identifiable information. The crawl-time filter, step by step:

The diagram shows six kinds of page leaving the index at crawl time, pages with helpful information going on to the index Perplexity serves to users, and four labels whose use Perplexity has not published.

The post also lists research planning among the workflows Decider fits, estimating how complex a question is before search begins so simple questions use fewer resources, without saying Perplexity’s own search runs it.

Perplexity did not publish the threshold its crawler applies, how many pages the filter removes, when filtering began or which Decider version runs in the crawler. The post says the right decision threshold depends on the application.

Why it happened

Perplexity states its reasons in its 12 November 2026 Decider post: many crawled pages are of no use to anyone, and Decider judges them faster and at lower cost than a large language model acting as a judge. The single pass that lets the crawler judge and label a page at once arrived in steps.

Decider v1.1 raised its Decision Index benchmark score from 56.4 to 61.56, its Hugging Face model card says. Decider v1.2 then read a whole request in one sequence, where v1.1 had answered one question at a time.

On 19 October 2026 Perplexity opened the Decisions API with v1.2, which it said answers multiple questions in one pass with up to five times faster inference. The quality filter sits in front of the index Perplexity ranks with its own engine.

What it changes for GEO

Brands whose CDN or firewall shows Perplexity’s crawler a challenge page lose that page from the Perplexity index. Perplexity names anti-bot walls and CAPTCHAs among the pages Decider filters out, so a page the crawler meets behind a wall counts as low quality. Owners often don’t know such a block exists, so the access check comes before any content budget.

Brands that paid for SEO filler lose on Perplexity too. Pages written around a keyword with no helpful information, and pages holding only links, are what Decider removes at crawl time, before ranking. Search groundwork still helps because retrieval begins with an index, but a retainer that delivers pages written around keywords with nothing a buyer can use now buys pages Perplexity filters out, and a page outside the index can’t be retrieved or cited.

AI engines judge a brand by how many trusted sources mention it and how widely. Decider filters spam pages out before they reach Perplexity’s index, so brands that bought mentions on spam sites lose those mentions on Perplexity.

A brand’s own site is the one source it controls, so the work goes there. The brands that gain publish comparisons and service pages in the buyer’s words, each answering a question in its first lines. 34.8% of the businesses in Agenzy’s 2026 source study did not have their own website among the 30 sources AI engines cited most for their own category (n=138 business projects).

Decider’s verdict belongs to Perplexity alone. A page Google indexes says nothing about its place in Perplexity’s index, so a brand’s position is judged engine by engine, since Perplexity works differently from ChatGPT. With no published threshold and no count of removed pages, any claim about the exact cutoff is an untested guess.

What to do

The head of marketing asks whoever runs the site’s CDN and firewall to confirm that PerplexityBot receives each page’s content and no challenge page, using the user-agent strings in the AI crawlers list. The CDN manager and the content team then work in this order:

• Read the CDN security log for PerplexityBot and confirm its requests show as allowed.
• List the pages that hold only links, or that were written around a keyword with nothing a buyer could use.
• Rewrite each listed page to answer one buyer question in its first lines, or remove it.

Perplexity’s Decider post names no way to look up a single page’s status in its index, so the check is twofold: the security log lists PerplexityBot as allowed, and Perplexity’s answers to the buyer’s questions cite the rewritten pages.