How to Benchmark Website Performance Against Competitors in AI Search

This benchmark measures website performance, not brand performance. A brand benchmark asks how often answers name you. A website benchmark asks which pages AI engines actually read and quote, yours or your rivals'. The two metrics at its core, Citation Share and Citation Rate, are defined with formulas in our GEO Metrics Framework. For brand baselines over time and the smallest gap you can trust, see how to benchmark AI search performance.

Why pages matter so much: engines now go looking for them. In a 2026 study of about 4,000 prompts, Nectiv found that ChatGPT ran 7.61 searches per prompt, up from 2.17, and 64% of them used the site: operator to search specific official sites. When an engine searches your rival's domain directly, the pages it finds there set the bar for your own website performance.

What a Website Performance Benchmark Compares

Five numbers make up a website performance benchmark. All but the last come from the same set of AI answers, so one round of prompt runs fills in every rival at once.

Citation Share is the number to lead with. It's the site-level cousin of share of voice, and Microsoft uses the same idea in Bing Webmaster Tools, where Citation Share is "the percentage of citations attributed to your site out of all citations shown across all sites for that same grounding query." The framework's Citation Share entry has the full formula and a worked example. Our AI citation glossary entry explains how a citation differs from a mention.

Collect the Citation Data

You can gather the data for a website performance benchmark by hand. It takes a few hours the first time and less after that.

1. Pick the domains to compare. Take two to four direct rivals. Then add the third-party sites that keep showing up as sources in your category, such as review sites, since they compete with you for the same slots.
2. Write 30 unbranded buyer prompts. Mix problem questions, "best tool for" questions and comparisons. Leave every brand name out. The free AI Visibility Prompt Kit drafts 30 from your category and a rival.
3. Run each prompt at least twice per engine. Use a clean session from the same location every time. AI answers vary between runs, so one pass isn't a benchmark.
4. Copy every cited URL. In ChatGPT, OpenAI says you can "select Sources, when available, to view cited sources and other relevant links." Perplexity numbers its sources under each answer. Paste one row per URL: date, engine, prompt, run, URL, domain.
5. Tag each URL with a page type. Use a short fixed list: homepage, pricing, comparison or alternatives, integrations and docs, how-to guide, listicle, review page, forum thread.

Add the first-party reports for your own site

Google and Microsoft report part of this for your domain, which helps you check your panel against a larger sample. Bing's AI Performance report shows total citations, cited pages and "citation counts for specific URLs from your site" across Copilot and AI summaries in Bing. Google's Search Console report shows how often your URLs appeared in AI Overviews and AI Mode. Both cover your own site. Microsoft even notes that Citation Share "does not expose competitor domains," so for rivals' pages the prompt panel is your source.

When a tracker saves time

Trackers run the prompts on a schedule and collect the cited URLs for you. Peec AI's Gap Analysis "highlights sources where your competitors appear but your brand does not," and its Starter plan was $95 a month for 50 prompts on its pricing page, as listed on 28 September 2026. Ahrefs' Brand Radar includes a Cited pages report, with its AI Visibility Index listed at $199 a month and custom prompts from $50 a month on the same date. Google's reminder applies to all of them: "No third-party tool has access to our internal ranking or AI systems." Trackers watch answers the way your panel does, at larger scale.

The Domain Citation Ledger: A Worked Example

The Domain Citation Ledger is Rankbox's one-sheet format for benchmarking website performance: one row per domain, then one row per page type for you and your strongest rival. Here it is for Plannora, a made-up project management tool, against a made-up rival, Loopcraft, a third made-up brand, Taskwell, and a made-up review site, stackreview.co. The panel is 30 prompts in ChatGPT and Perplexity, run twice each: 120 answers citing 744 URLs in total. All numbers are illustrative.

Sheet 1: citation share and rate by domain

Loopcraft's site wins more than three times Plannora's share of sources. The review site is linked in more answers than any vendor, which makes it a target for Plannora's outreach, not just a rival.

Sheet 2: which page types win the citations

The ledger explains the gap. More than half of Loopcraft's citations (55 of 97) come from two page types Plannora doesn't have at all. Plannora's how-to guides hold their own, with 18 citations to Loopcraft's 16. Its pricing page is the odd one out: it exists but earns 3 citations to Loopcraft's 17. That's a flag for a technical check, covered below.

Reading the ledger

1. Taskwell's number is an access story. All 44 of its citations came from ChatGPT and none from Perplexity. Its robots.txt blocks PerplexityBot, so Plannora compares with Taskwell in ChatGPT only, where Taskwell holds 44 of 402 cited URLs, or 10.9%.
2. Missing page types come first. Comparison pages and integration docs are where the citations are and where Plannora has nothing.
3. Existing pages that underperform come second. The pricing page gets a technical check before any rewrite.

Find the Content Gaps

Content-gap analysis is where a website performance benchmark turns into a to-do list. It asks a simple question: which of your rivals' cited pages have no counterpart on your site? Sheet 2 gives you the page types. Now go one level down, to the URLs.

1. List each rival's cited URLs with their citation counts, sorted from most to least.
2. Match each one to your closest page. Write the URL, or "none."
3. Score the gaps. A rival page with many citations and no match on your site goes to the top.
4. Check which prompts pulled each rival page. That tells you what the new page must answer in its first lines.
5. Re-check by engine. Different engines reward different page types.

That last step matters because engines differ. In Peec's study of over 1 million citations, published in February 2026, listicles made up just under a fifth of the URLs ChatGPT used, and 52% of them landed in its top citation bucket. Google AI Mode showed no favourite page type and tended "to prefer official sources, such as brand websites and their product and category pages." A gap that costs you in one engine may not matter in another.

Some gaps sit on other sites. If review sites or forum threads win a big share, the answer is presence there: a listing, an update to wrong facts, or a genuine reply. For the writing side of new pages, see our guide to optimizing content for AI search.

Compare Crawl Access and Technical Health

Before you blame content, check whether both sites can be read. Website performance in AI search starts with access, and a rival's robots.txt is public.

Crawl access

Each engine's search crawler decides whether a site's text can appear. OpenAI says sites that opt out of OAI-SearchBot "will not be shown in ChatGPT search answers, though can still appear as navigational links." Perplexity says PerplexityBot is "designed to surface and link websites in search results on Perplexity." Anthropic says turning off Claude-SearchBot "may reduce your site's visibility and accuracy in user search results."

How often do sites block? Rankbox's AI Bot Crawler Census of 21,353 sites found: "OAI-SearchBot is blocked by 28.78% of news sites, 3.54% of retailers and 0.81% of the Fortune 500." Among software and SaaS sites, the rate was 1.28%. The census adds: "Only 14.1% of e-commerce sites fully block at least one AI bot, against 62.2% of news sites. The Fortune 500 sits at 3.8%, software and SaaS companies at 7.0%." It also found: "A quarter of e-commerce sites (25.1%) wouldn't serve robots.txt to the census crawler at all." For those sites, a firewall may make the call before robots.txt does.

The technical check table

Run the same checks on your site and on each rival. Most can be done from the outside.

The raw-HTML check catches a common problem. Vercel's crawler analysis found that "none of the major AI crawlers currently render JavaScript," naming OpenAI's crawlers, ClaudeBot and PerplexityBot, while Gemini renders through Googlebot. In the worked example, this is where Plannora's pricing problem turns up: its price table loads by script, so the raw HTML has no prices. Loopcraft's prices sit in plain HTML. The fix comes before any rewrite.

For a quick scan of one URL, the free AI search readiness check tests 12 AI-readiness signals. For site-wide crawler and CDN fixes, see our guide to optimizing a website for ChatGPT and Perplexity.

Where Rankbox Fits

Rankbox doesn't track AI citations today, so it won't benchmark website performance for you. Run the panel by hand or with one of the trackers above. Where Rankbox helps is closing the gaps the ledger finds. Answer-Space Research maps the questions buyers ask AI in your category, so your new comparison and docs pages answer the prompts rivals win. The Citation-Ready Writer writes source-backed articles that reach your site through Rankbox's API. The Business plan is $49.50 a month with a 7-day trial: see pricing.