Rendix
[01]

Product

Technical SEO for AI search.

Vladimir Bazalenko, Founder, RendixUpdated Oct 9, 2026

Technical SEO for AI search is the work that lets search and AI crawlers reach, read and index your pages, because AI answers are built from what those indexes hold.

Rendix audits your site the way the crawlers see it: what each search, user and training bot gets, the text that is there without JavaScript, and the index status Google and Bing report. Then it shows which of the visits were really those bots.

7-day free trial. Cancel anytime.

Who reads rendix.ioSep 9 to Oct 8
Search index bots1,465 verified · 671 posing

They fill the indexes AI answers draw from. Blocking one keeps your pages out of that engine's search.

  • OAI-SearchBotOpenAIChatGPT search839382 posing
  • PerplexityBotPerplexityPerplexity answers32197 posing
  • GooglebotGoogleSearch, AI Overviews, AI Mode20022 posing
  • BingbotMicrosoftBing and Copilot39458 posing
  • Claude-SearchBotAnthropicClaude search012 posing
Sep 9 to Oct 8. Verified: from an address the operator publishes. robots.txt allows each crawler here, and each got a 200 on Oct 9.Who is really crawling
[02]

Definition

What is technical SEO for AI search?

The same base as SEO: AI answers are built from what search indexes hold and rank. Three facts behind it:

The answer comes from an index

AI search finds pages through search indexes. Google's AI features only use pages in Google's index.

87 % of ChatGPT search's citations matched Bing's top results in Seer's test; 56 % matched Google's.

100 queries, 500+ citations, Feb 2025; a later replication found a lower match once ChatGPT ran several searches per question. Google: a page must be indexed and allowed a snippet to appear in AI Overviews or AI Mode.

Source: Seer Interactive, Feb 2025 · Google Search Central, Dec 2025

Search rank decides what gets read

Of the pages an engine's search returns, the top one is cited about four times as often as the tenth.

58 % of ChatGPT's first search results were cited in its answer, against 14 % of the results at position 10.

AirOps and Kevin Indig, 16,851 queries run three times in ChatGPT, Apr 2026. Observational, ChatGPT only. Google puts it plainly: optimizing for generative AI search is still SEO.

Source: AirOps, Apr 2026 · Google Search Central, Jul 2026

AI crawlers read the HTML only

Only Google's and Apple's crawlers ran JavaScript in Vercel's study, so text drawn by scripts is missing for the rest.

11.5 % of ChatGPT's crawler requests fetched JavaScript files, and none of them ran; Claude's 23.8 %.

Crawler traffic on Vercel's network, Dec 2024. Fetching a script file is not running it. A full bar is 50 %.

Source: Vercel and MERJ, Dec 2024

[03]

Method

How does Rendix audit your site?

The checks of a technical SEO audit, made as the crawlers make their requests, then read against Google's and Bing's own index data.

  1. Collect the pages you declare

    Up to 400 pages from your sitemaps (ten files, child sitemaps two levels deep) and llms.txt, the homepage first.

  2. Knock as each crawler

    The homepage is requested as ten search, user and training bots, and robots.txt is read for each. Pages are then fetched as the first search crawler the site lets in, because bot management often refuses unknown clients.

  3. Read what a crawler gets

    Status, redirects, canonical, noindex and nosnippet, the text present without JavaScript, titles, headings and hreflang. A page the site refused is marked not measured, never broken.

  4. Ask the indexes

    With Search Console connected, each page's index status comes from Google's URL Inspection. With a Bing key, the last Bingbot crawl of each page; Bing publishes no index flag.

  5. Score what is open

    Each open finding costs points by severity: critical 25, high 15, medium 8, low 3. Probe limits and hygiene notes such as schema are listed and never scored.

  6. Fix, re-run, inspect

    Run the audit again after a fix, inspect a single URL live, or export every page and finding as CSV.

SEO base, AI layerfive layers, the base first
hover a layer
What each stage checksfrom the checks catalog
StageIt asksChecksRead from
ReachableCan the crawlers get the page?robots.txt or CDN blocks a search bot, 4xx and 5xx, redirect chains, stale sitemap URLsCrawler requests
ReadableDoes a crawler get the text?content only after JavaScript, nosnippet or a tight max-snippet, thin text, slow first byteCrawler requests
StructuredCan it tell what the page is?missing title or description, duplicate titles, heading structure, hreflang, orphan pagesThe page's HTML
IndexedIs it in Google's index?not indexed, Google chose another canonical, never crawled by Bingbot (a note)Search Console, Bing
CitedDoes a cited page still work?a page an AI answer cites returns an errorRendix scans
Not checked on purpose: Core Web Vitals, mobile layout, URL cosmetics and compression. The AI crawlers above do not render a page, and Google ties speed to how much it crawls, which the audit reads through server response time.
[04]

Crawlers

Who is really crawling your site?

Check the address a visit came from. Any client can call itself GPTBot; OpenAI, Anthropic, Perplexity, Google and Microsoft publish the addresses their bots use, and only a visit from one of them is the bot.

On rendix.io, 41.2 % of the 2,938 visits that used a crawler's name in 30 days came from somewhere else: 89.6 % of ClaudeBot's, 86.0 % of PerplexityBot's, 9.9 % of Googlebot's. They can be scrapers or other tools borrowing the name; the log cannot tell which.

A name misleads the other way too. A large retail site Rendix audits answered 429 to a browser and to an honest audit client, and 200 to every crawler's name. So the audit fetches pages as the search crawler the site lets in, and a page it refuses counts as not measured.

Source: Rendix server logs, rendix.io · OpenAI bots docs

Visits that only used the namerendix.io · Sep 9 to Oct 8
Search index bots
OAI-SearchBot382 of 1,221 visits31.3 %
PerplexityBot197 of 229 visits86.0 %
Googlebot22 of 222 visits9.9 %
Bingbot58 of 454 visits12.8 %
Claude-SearchBot12 of 12 visits100.0 %
User-triggered fetches
ChatGPT-User53 of 105 visits50.5 %
Perplexity-User12 of 12 visits100.0 %
Claude-User8 of 17 visits47.1 %
Training crawlers
GPTBot275 of 454 visits60.6 %
ClaudeBot190 of 212 visits89.6 %
Share of each crawler's visits from outside its operator's published addresses, Sep 9 to Oct 8, 2026. 71 visits during Rendix's own audit runs left out.
[05]

In Rendix

What does Rendix show in Foundation?

The audit, the bots' access and the index data in one place, re-run after every fix.

Findings

Open problems ranked by severity, each with the pages it covers, the fix and the points clearing it adds back.

Foundation · Findings

Pages

Every audited page with its status, what the crawler got and the flags on it.

Foundation · Pages

Access

Each bot's robots.txt rule and answer, its verified visits once your server logs are connected, and your Search Console and Bing connections.

Foundation · Access

History

Every run's score and findings, so a fix shows as a finding that cleared.

Foundation · History
In development
  • AI crawler visits per pageVerified bot visits from your logsSoon
  • Page briefsWhat a page answers, what it missesSoon
  • An SEO sectionNext to FoundationSoon

The fixes in your plan

Technical findings that keep a page from being read join the publishing plan, next to the pages to write.

Actions

The pages the engines cite

Which of your pages, and which of your competitors', the answers cite.

Citations
[06]

Plans

What you get

Straight from the pricing table: only what ships today.

What you getEvery plan
  • Technical site auditCrawlability, indexation, sitemap, canonicalsIncluded
  • Re-run and inspect any URLIncluded
  • Search Console and Bing index statusConnect your own propertiesIncluded
  • CSV exportIncluded
Compare plans
[07]

Questions

/ Updated Oct 9, 2026

Questions about technical SEO for AI

Need more help? team@rendix.io

The work that lets search and AI crawlers reach, read and index your pages, so AI answers can use them.

It is the technical SEO base seen from the AI side: which bots may fetch a page, what text they get without JavaScript, and whether Google's and Bing's indexes hold it. Rendix audits all three and checks which crawler visits are real.

It builds on SEO: Google says optimizing for its generative AI features is still SEO.

The names GEO and AEO describe the layer on top, being named and cited inside the answer. That layer stands on pages the engines' indexes hold and rank: in ChatGPT, the first search result was cited 58 % of the time, the tenth 14 %.

Source: Google Search Central, Jul 2026 · AirOps, Apr 2026

Request your pages as each crawler, read robots.txt for each, then check who actually visits.

A request under a bot's name shows what your server answers that name. It does not show who visits: on rendix.io, 41.2 % of the visits using a crawler's name came from addresses the operators do not publish. Rendix runs both checks and adds Google's and Bing's index data.

Source: Rendix server logs, rendix.io

Blocking GPTBot keeps your pages out of OpenAI's model training and leaves ChatGPT search as it is.

ChatGPT search uses OAI-SearchBot, and blocking that one removes the site from ChatGPT search. Anthropic splits the same way (ClaudeBot trains, Claude-SearchBot indexes). Google-Extended is a robots.txt token with no crawler of its own: it controls Gemini training and grounding and does not affect Google Search.

Source: OpenAI bots docs · Anthropic crawler docs · Google crawlers docs

Google's and Apple's do. In Vercel's study the other AI crawlers did not.

ChatGPT's and Claude's crawlers fetched JavaScript files without running them, so text that appears only after scripts run is missing for them. Rendix flags pages whose content needs JavaScript and reads every page the way a non-rendering crawler gets it.

Source: Vercel and MERJ, Dec 2024

It changes how much Google crawls. Rendix knows of no study that shows it moves AI citations much.

Google lowers its crawl rate when a server slows down, returns errors or answers 429. Rendix measures server response time and refused requests, and leaves Core Web Vitals to classic SEO tools, since the AI crawlers do not render the page.

Source: Google Search Central, Jul 2026

Because only they know whether your pages are in the index.

Search Console's URL Inspection returns Google's verdict for each page, within 2,000 inspections a day per property; Bing's API returns the last Bingbot crawl but no index flag. Rendix reads both with your own access and never shows a page as indexed without a source.

Source: Search Console API docs

[08]

Get started

/ Free 7-day trial

See where you stand Get the plan

One scan shows your place in AI answers – and the first moves to climb. Free, no account needed.