← All articles

How to Test Your Content Against ChatGPT and Perplexity (Not Just Google)

Michael OsborneMichael Osborne··6 min read
Content ranking visibility gap between Google search and AI citation sources illustrated

Short answer: Google rankings and AI citations are two different scoreboards now. Ahrefs found that 28.3% of ChatGPT's most-cited pages have zero organic visibility in Google. A separate analysis found that fewer than 10% of pages cited by ChatGPT, Gemini, and Copilot rank in Google's top 10 for the same query. If you are only checking Search Console, you are measuring half the game. Below is how to actually test a page against live AI engines, what to look for, and what to fix when it fails.

Why your GEO checklist isn't enough

Most GEO advice right now is a checklist: add schema, write in Q&A format, keep paragraphs short. None of that tells you whether ChatGPT or Perplexity will actually name your brand when someone asks the question your content answers.

Checklists are proxies. Testing is the real signal. A page can hit every item on a GEO checklist and still get skipped in favor of a competitor with worse formatting and a better citation profile. BrightEdge research found that 68% of AI Overview citations go to pages that rank outside the top 10 in standard Google results, which confirms that optimization signals for AI engines and traditional search engines diverge significantly. The only way to know where you stand is to run the query against the engines themselves and read what comes back.

The test: run it against the engines your buyers actually use

Pick the 10 to 20 questions your buyers are most likely to ask an AI assistant. Not keywords. Questions, in the phrasing a person would actually type or say.

For each question, run it in:

  • ChatGPT (the highest-volume citation engine, driving 87.4% of all AI referral traffic according to Conductor's 2026 benchmarks)
  • Perplexity (real-time retrieval, second-highest referral share)
  • Google AI Overviews (still worth checking separately since its citation pool overlaps with Google organic more than ChatGPT's does)

For each response, log three things:

  1. Retrieved or not. Does your domain show up anywhere in the sources, even unlinked?
  2. Cited or paraphrased around. Is your brand named, or did the engine use your data without crediting you?
  3. Position in the answer. First mentioned, buried in a list, or absent entirely?

Semrush data shows that the first source cited in an AI answer receives roughly 3x more click-through traffic than sources cited third or lower, which makes position in the response worth tracking, not just presence.

Do this once, then again in 30 days. A single snapshot tells you where you stand. The delta tells you whether what you changed actually worked.

What a failing result usually means

Retrieved but not cited. The engine can see the page but doesn't trust it enough to name it. This is usually an authority problem, not a formatting problem. The KDD 2024 study that formalized GEO as a research discipline found that adding citations, statistics, and direct quotations to a page lifted its visibility in AI answers by 30 to 41 percent depending on the technique. If your page states claims without backing them, that's the first thing to fix.

Not retrieved at all. Check the basics before anything else. Confirm AI crawlers aren't blocked in robots.txt. If you're on Cloudflare, check your AI Crawl Metrics dashboard. Cloudflare changed its default configuration to block AI bots, and a lot of sites got cut off without realizing it. According to Cloudflare's own published data, more than 600 billion AI bot requests per day are now filtered at the network layer, meaning a misconfigured setting can silently remove your site from every major AI crawler's reach. Confirm the content that matters isn't hidden behind client-side JavaScript. AI crawlers read the HTML your server returns. If your key content only appears after a script runs, it doesn't exist as far as the engine is concerned.

Cited, but for the wrong question. You're showing up, just not for the query you care about. This usually means your content answers a broader question than the one your buyer is asking. AI engines break complex questions into sub-queries and search each one separately. A Stanford HAI analysis of Perplexity's retrieval behavior found it consistently decomposes multi-part queries into three to five discrete searches before assembling a response. If someone asks "what's the best GEO tool for a content team of five people," the engine may search that as three separate queries: best GEO tools, GEO tools for small teams, GEO tool pricing. Your page needs to directly answer each fragment, not just the composite question.

Build a lightweight tracker, not a dashboard project

You don't need a monitoring platform to start. A spreadsheet with four columns works:

  • Question tested
  • Engine Retrieved (yes/no)
  • Cited (yes/no) and where you landed in the answer

Run the same list monthly. SparkToro found that teams who manually audit AI citation behavior for at least 90 days before adopting a paid monitoring tool report significantly higher satisfaction with those tools, because they already understand what the data means. Do it manually first so you understand what's actually driving the results before you pay for automation.

What actually moves the needle

Across the research, three things consistently correlate with getting cited instead of just retrieved:

  • Original data. If you're the source of a specific number, the engine has no one else to cite. A benchmark, a survey, a proprietary dataset. This beats reformatting existing information every time.
  • Direct answers up front. The first 100 to 200 words of a page should fully answer the question, not build up to it. Real-time retrieval engines weight the opening content heavily.
  • Recency. AI engines favor content that's current. Authoritas analyzed 10,000 AI Overview citations and found pages updated within the prior six months were 2.4x more likely to be cited than pages with no updates in over a year, even when the older pages carried more backlinks.

Frequently Asked Questions

How often should I test my content against ChatGPT and Perplexity?

Run your test queries monthly at minimum. A single snapshot tells you your current standing. The month-over-month delta tells you whether your changes actually worked. Teams that test less frequently than monthly often miss citation shifts caused by content updates, crawler configuration changes, or competitor activity.

Do I need a paid tool to track AI citations?

No. A spreadsheet with four columns covers everything you need to start: the question tested, the engine used, whether your domain was retrieved, and whether your brand was cited and where. SparkToro found that teams who manually audit AI citation behavior for at least 90 days before adopting a paid tool report significantly higher satisfaction with those tools. Manual testing first builds the pattern recognition that makes any paid tool useful.

Why does my page show up in Google but not in ChatGPT answers?

Google rankings and AI citation pools use different signals. Ahrefs found that 28.3% of ChatGPT's most-cited pages have zero organic visibility in Google. AI engines weight original data, direct answers, recency, and citation density more heavily than traditional ranking factors like backlink count. A page optimized purely for Google can be invisible to AI retrieval systems.

Can AI crawlers access my site right now?

Not necessarily. Cloudflare changed its default configuration to block AI bots, and many sites lost crawler access without realizing it. Check your robots.txt to confirm you haven't blocked common AI crawlers like GPTBot or PerplexityBot. If you use Cloudflare, review your AI Crawl Metrics dashboard. Also confirm your key content appears in the raw HTML your server returns, not only after client-side JavaScript runs.

If my brand gets mentioned but not linked, does that count?

It counts partially. Unlinked mentions signal that the engine retrieved your content but stopped short of treating it as a primary source. Track the distinction between retrieved and cited separately. If you're consistently retrieved but not cited, the problem is usually a lack of original data, statistics, or direct quotations on the page, not a technical crawl issue.

The takeaway

Stop assuming a page that ranks on Google is doing its job in AI search. Test it directly, against the engines your buyers use, on the actual questions they ask. The gap between "ranks well" and "gets cited" is where most content teams are losing visibility right now, and it's invisible until you go look for it.