How to Optimize Your Website for ChatGPT and Perplexity

That split shows up in the data. In Ahrefs' August 2025 study of 15,000 prompts, 28.6% of the pages Perplexity cited ranked in Google's top 10 for the query. For ChatGPT, Gemini and Copilot the figure hovered around 8%. Same web, different pickers.

The rules also moved this month. From 15 September 2026, Cloudflare blocks "Agent" bots by default on ad-carrying pages of newly added domains, and its own example of an agent is ChatGPT-User. A site can pass every content check and still lose live fetches from ChatGPT and Perplexity to a CDN default.

This guide gives you one checklist for ChatGPT and Perplexity, with a way to verify each item yourself in a minute, then maps the six places where they split. For crawler internals, see our technical guides to ChatGPT SEO and Perplexity SEO.

How ChatGPT and Perplexity Reach Your Site

Both engines answer from the live web. They build their pool of pages in different ways, and that one fact explains most of the checklist below.

ChatGPT borrows results and runs its own crawler

ChatGPT decides when a question needs a search. When it searches, OpenAI's help page says it "typically rewrites your query into one or more targeted queries" and sends them to partner search providers. The page points to Microsoft's and Shopify's privacy policies for those providers.

OpenAI also runs its own crawler. Its crawler docs say OAI-SearchBot "is used to surface websites in search results in ChatGPT's search features." So a page reaches ChatGPT through two doors: a provider's index, such as Bing, and OpenAI's own crawl.

The searches themselves have changed shape. Nectiv's 2026 study of about 4,000 prompts found ChatGPT running 7.61 searches per prompt, up from 2.17, with the site: operator in 64% of them. "Official" and "gov" were among the words it added most. That is query fan-out aimed straight at official domains.

Perplexity runs its own index and ranks passages

Perplexity started out on third-party search APIs, then built its own index. Its September 2025 architecture write-up says the index "tracks over 200 billion unique URLs." A model predicts when each URL needs a recrawl, based on its importance and how often it changes.

Pages are then split into "self-contained spans, each of which can be individually retrieved and ranked." Prefilters drop "clearly non-responsive or stale content" before ranking starts. So being in Bing's index doesn't help Perplexity directly. What counts is whether PerplexityBot can fetch you, how often your pages change, and whether each section stands on its own.

ChatGPT and Perplexity side by side

The rows come from OpenAI's crawler docs and publisher FAQ, Perplexity's crawler docs and help center, IndexNow's list of engines and Cloudflare's referrer list, all checked in September 2026.

The Two-Engine Site Checklist for ChatGPT and Perplexity

We call this the two-engine site checklist. Each row is one check, what each engine needs, whether the fix is the same, and how to confirm it yourself. Work from the top down. A failure high on the list blocks every row below it.

Ten of the fifteen rows are the same fix for ChatGPT and Perplexity. Do those first, because each one counts twice.

The five verification commands

Every command below runs in a normal terminal on macOS or Linux. Swap in your own domain, page and a sentence you expect to find.

1. Read what robots.txt tells each crawler. A Disallow: / under User-agent: * blocks any crawler that has no group of its own.

curl -s https://yoursite.com/robots.txt | grep -i -A3 -E "^user-agent: *(\*|oai-searchbot|perplexitybot|gptbot)"

1. Fetch a page the way each crawler asks for it. The loop uses the user-agent strings from each vendor's docs (OpenAI notes its version number may change). The first line of each pair is the status code and time to first byte. The second is how many lines contain your key sentence. Zero means the sentence isn't in the raw HTML: JavaScript adds it, the wording differs, or your firewall served a challenge page.

for ua in \
  "Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/131.0.0.0 Safari/537.36; compatible; OAI-SearchBot/1.4; +https://openai.com/searchbot" \
  "Mozilla/5.0 AppleWebKit/537.36 (KHTML, like Gecko; compatible; PerplexityBot/1.0; +https://perplexity.ai/perplexitybot)"; do
  curl -s -o /dev/null -w "%{http_code} %{time_starttransfer}s\n" -A "$ua" https://yoursite.com/pricing
  curl -s -A "$ua" https://yoursite.com/pricing | grep -c "Plans start at"
done

This tests your server and any user-agent rules. It can't copy the bots' IP addresses, so a CDN may treat your test differently from the real crawler. Command 3 shows what the real bots got.

1. Count the status codes each AI bot received. This assumes the common "combined" log format, where the ninth field is the status code.

for bot in OAI-SearchBot ChatGPT-User PerplexityBot Perplexity-User; do
  echo "$bot"; grep "$bot" access.log | awk '{print $9}' | sort | uniq -c
done

1. Check that a bot is who it claims to be. User agents are easy to fake. Both vendors publish their IP ranges, so test a suspect IP against the list. Replace the example IP with one from your logs, and swap the URL for https://openai.com/searchbot.json, chatgpt-user.json or perplexity-user.json to check the other bots. It prints True or False.

IP=203.0.113.7
curl -sL https://www.perplexity.com/perplexitybot.json | python3 -c "
import sys, json, ipaddress
ip = ipaddress.ip_address('$IP')
nets = [p.get('ipv4Prefix') or p.get('ipv6Prefix') for p in json.load(sys.stdin)['prefixes']]
print(any(ip in ipaddress.ip_network(n) for n in nets))"

1. Ping IndexNow after you publish. Host your key file first. A 200 means the URL was received. IndexNow's FAQ says a first request may return 202 while your key is checked.

curl -s -o /dev/null -w "%{http_code}\n" "https://api.indexnow.org/indexnow?url=https://yoursite.com/pricing&key=YOUR_KEY"

If you'd rather not use a terminal, our free AI crawler log analyzer reads a day of access logs in your browser and shows each bot's hits and errors.

Why the firewall rows matter more than robots.txt

Rows 3 and 4 are the easiest to fail without noticing. OpenAI says "robots.txt rules may not apply" to ChatGPT-User, and Perplexity says Perplexity-User "generally ignores robots.txt rules." Both are fetching a page because a person asked a question. So robots.txt can't reliably block them, and it can't help them past a firewall either.

Cloudflare's AI Crawl Control list files OAI-SearchBot and PerplexityBot as AI Search, and ChatGPT-User and Perplexity-User as AI Assistant. Under its new defaults, search stays allowed and agents are blocked on ad-carrying pages of new domains. Perplexity's crawler docs include ready-made Cloudflare and AWS WAF rules that match its user agents and IP ranges. OpenAI's help page asks you to confirm that your host or CDN "allows traffic from OpenAI's published searchbot IP addresses."

The Divergence Map: Where ChatGPT and Perplexity Split

Most of the checklist is one fix, applied once. In six places, ChatGPT and Perplexity want different things. We call them the divergence map: the spots where a change helps one engine and does little or nothing for the other.

1. Bing and IndexNow only help one engine

Microsoft is one of ChatGPT's search providers, so a page Bing has indexed gives ChatGPT a second way to find it. IndexNow tells Bing about a new or changed URL the moment you publish, and Bing's sitemap guidance calls an accurate lastmod "a key signal" for recrawling. Neither step touches Perplexity. It isn't on IndexNow's list of participating engines as of September 2026, and it documents no URL submission tool.

Perplexity learns your rhythm instead. Its write-up says the recrawl model lets it defer visits "for sites with routine publication and refresh cadences" to when they're "likely to be most useful." Publish on a steady schedule, and link new pages from pages that change often, such as your blog index or changelog.

2. Rankings carry over to Perplexity more than to ChatGPT

Ahrefs found Perplexity's citations overlap with Google's top 10 far more than ChatGPT's do. Perplexity doesn't read Google's results. It runs its own index, yet its picks land closer to classic search. Strong SEO pages have a head start there.

For ChatGPT and the other assistants, Ahrefs found more than 80% of citations came from pages that don't rank at all for the original prompt. That fits what OpenAI describes: ChatGPT rewrites the question into its own targeted searches. With site: searches now common, the vendor's own page for that exact fact gets a direct shot.

3. Only ChatGPT asks you to make a training call

OpenAI keeps search and training apart: "Each setting is independent of the others." You can block GPTBot and still be cited. Perplexity's docs say PerplexityBot "is not used to crawl content for AI foundation models," so there's no second Perplexity bot to decide on.

This matters at your CDN too. Under Cloudflare's new rules, a crawler that mixes search and training is blocked once you block training. Neither OAI-SearchBot nor PerplexityBot is a training crawler, so a training block leaves both search crawlers alone.

4. ChatGPT and Perplexity both reward fresh pages

Perplexity is explicit about freshness: stale content is filtered out before ranking. ChatGPT publishes no such rule. Yet in Ahrefs' July 2025 freshness study of 17 million citations, ChatGPT's cited pages averaged 958 days old, against 1,166 for Perplexity and 1,416 for Google's organic results. Ahrefs also found both engines appear to order their in-text references from newest to oldest.

The fix is the same for both: show a real published date and a real updated date, and change the date only when the content changes. For pricing and specs, refresh on every change and ping IndexNow.

5. A blocked page doesn't fully leave ChatGPT and Perplexity

OpenAI says a site blocked from OAI-SearchBot "can still appear as navigational links." Its publisher FAQ says to use a noindex meta tag to stop that, and the crawler must be allowed to fetch the page to see the tag.

Perplexity's robots.txt help page says that for a blocked page, "we may still index the domain, headline, and a brief factual summary." It doesn't document noindex handling. If you need Perplexity to stay away entirely, robots.txt alone won't do it. Blocking both bots at your firewall, by user agent and IP, is the stronger option.

6. Your analytics sees the two differently

OpenAI's FAQ says ChatGPT "automatically includes the UTM parameter utm_source=chatgpt.com in referral URLs." Google's channel definition groups ChatGPT, Gemini, DeepSeek, Copilot and Grok into GA4's AI Assistant channel. It doesn't name Perplexity, so check your Referral report for perplexity.ai and add it to a custom channel. Our guide to measuring GEO has a regex that covers both.

Pages ChatGPT and Perplexity Both Reach For

Access gets you into the pool. Page inventory decides whether you have something worth citing. Both ChatGPT and Perplexity pull from the same kinds of first-party pages, for slightly different reasons.

A few rules make these pages work for both engines:

• One question per page, or per clearly headed section. Perplexity ranks spans on their own, and ChatGPT's rewritten queries are targeted. A page that answers ten things gives both of them ten weak passages.
• Put comparisons and specs in real HTML tables. Perplexity's write-up says list- and table-heavy pages "may benefit from more formulaic parsing."
• Use plain, descriptive URLs and titles. "/pricing" and "/compare/loopcraft" say what the page answers before anyone opens it.
• Keep facts identical everywhere. A price that differs between your pricing page and a docs page gives either engine two answers to choose from.
• Make syndicated copies point home. In a 2025 Tow Center test of eight AI search tools, answers sometimes cited syndicated copies on Yahoo News or AOL instead of the original. Ask partners for a canonical tag that points back to your page.

How you write each section is its own craft. Our guide to optimizing content for AI search covers answer-first writing at the passage level. The about page also feeds your wider entity work, covered in our guide to optimizing your business for AI search.

A Worked Example: Auditing Plannora

Plannora is a made-up project management tool at plannora.io. Its team checks the site against both ChatGPT and Perplexity using the two-engine checklist. Rows 2, 7 and 8 only apply to ChatGPT, so ChatGPT has 15 rows to pass and Perplexity has 12.

The starting score: ChatGPT fails rows 5, 8, 11 and 12, so Plannora passes 11 of 15 (73%). Perplexity fails rows 4, 5, 11, 12 and 15, so it passes 7 of 12 (58%).

The fix order follows the arithmetic:

1. Fix the shared rows first. Rows 5, 11 and 12 fail for both engines. Three fixes clear six failures. ChatGPT rises to 14 of 15 (93%) and Perplexity to 10 of 12 (83%).
2. Fix the Perplexity-only rows next. Rows 4 and 15 take Perplexity to 12 of 12.
3. Finish with IndexNow. Row 8 takes ChatGPT to 15 of 15.
4. Re-run all five commands a week later to confirm each fix held.

A full pass doesn't guarantee a single citation. It removes every technical reason for ChatGPT and Perplexity not to cite you. The rest comes down to whether Plannora's pages answer the question better than the next site's.

What ChatGPT and Perplexity Don't Ask For

Some popular fixes don't appear anywhere in either vendor's documentation. Skip them until the checklist is green.

• llms.txt. Neither OpenAI's nor Perplexity's crawler docs mention reading your llms.txt file. Perplexity publishes one for its own help and developer docs, which is a different job. A file does no harm, but don't expect it to move citations.
• A submission console. Neither vendor documents a URL submission form or webmaster console. OpenAI's FAQ says: "Any public website can appear in ChatGPT search."
• Schema as a ranking switch. Neither vendor's crawler docs mention structured data. Keep schema accurate for Google and Bing, and make sure it matches the visible text.
• Blocking GPTBot to "protect" search. It changes nothing about ChatGPT search, in either direction.

One more thing: OpenAI's FAQ says its Atlas browser uses ARIA tags "to interpret page structure and interactive elements" when ChatGPT's agent works on a page. If you want agents to book demos or fill forms on your site, accessible markup now pulls double duty.

Where Rankbox Fits in the Two-Engine Checklist

Rankbox helps with row 12, the page inventory, not the plumbing. Its Answer-Space Research maps the questions your buyers ask ChatGPT, Perplexity and Google, and scores them by estimated volume, difficulty and intent. The Citation-Ready Writer researches the live web and writes source-backed articles of 2,000 to 3,500 words. Articles reach your site through Rankbox's API, which a developer wires in once.

Rankbox doesn't change your robots.txt, CDN or firewall, and it doesn't track AI citations today. For the access rows, use the free tools linked above. For tracking, see our round-ups of ChatGPT rank trackers and Perplexity SEO tools. The Business plan is $49.50 a month with a 7-day trial: see pricing.