The 10-minute AI search audit anyone can run.

Gartner projects that 25% of traditional search volume migrates to AI-assistant interfaces by the end of 2026. The Invisible 10 study Web Cited ran in May 2026 measured what that looks like in practice for one category: ten funded B2B compliance vendors tested against 600 LLM responses across ChatGPT, Claude, Gemini, and Perplexity returned zero citations of any of the ten named brands. The same 600 responses repeatedly named a different set of vendors: TalentLMS at 107 mentions, LogicGate at 73, AuditBoard at 69.

A buyer of one of those vendors is already running the prompts in this audit. The buyer sees the AI engine's answer before they see your homepage. The point of running the prompts yourself, in ten minutes, is that you see what the buyer sees.

This is not a measurement program. A measurement program uses multi-trial sampling, region controls, and confidence intervals. This is a screening question: yes or no, are we invisible? For most B2B brands the answer is yes, and that is the moment to start fixing the right thing instead of guessing.

• A laptop or phone with a browser.
• Five specific prompts your buyers actually use. Suggested examples are below if you do not have a list yet.
• Free-tier access to ChatGPT, Claude, Perplexity, and Gemini. No paid accounts needed.
• A note-taking surface. Text editor, paper, or a spreadsheet works.

Optional: Grok and DeepSeek for the full six-engine set Web Cited's SXO Audit covers. The four-engine version is enough to answer the yes-or-no question.

Step 1. Write down five buyer prompts (2 minutes)

These have to be prompts a real buyer would type. The most common mistake is writing the prompt in your category's internal language instead of the buyer's plain-English language. "Best GRC platform" is internal. "Compliance software for a 200-person healthcare company" is what the buyer types.

Three patterns that tend to produce useful prompts:

• Best [category] for [specific buyer situation]. Example: best business checking account for a twelve-person agency.
• How do I [job to be done]. Example: how do I run a SOC 2 audit at a Series B startup.
• Compare [your category] tools for [use case]. Example: compare e-signature tools for a real-estate brokerage.

Five prompts is enough variance to surface a real pattern without making the audit take twenty minutes.

Step 2. Open the four engines (1 minute)

• chatgpt.com
• claude.ai
• perplexity.ai
• gemini.google.com

One browser tab per engine. Free tiers are fine. If you have a paid ChatGPT or Claude account, log out first, or run the audit in a private window. Logged-in answers can include memory of prior conversations, which makes the audit non-reproducible.

Step 3. Run each prompt across each engine (5 minutes)

Five prompts times four engines is twenty trials. About fifteen seconds per trial, including reading the response.

For each trial, note:

• Did your brand name appear in the response? Yes or no.
• Which competitors appeared by name? List them.

One row per prompt-engine pair, twenty rows total. A small spreadsheet works well; so does a piece of paper with a grid. If you want a richer audit, note any third-party domains the engine cited inline; that becomes useful in step five.

Step 4. Score the audit (1 minute)

Two numbers are enough.

• Coverage: out of four engines, how many cited your brand in at least one of the five prompts?
• Density: across twenty trials, how many cited your brand at all?

Coverage answers "are we on the AI search map?" Density answers "how often?"

Step 5. Bonus prompts (1 minute, optional)

Three direct-name tests are worth running on at least one engine each:

• Best [your category] tools 2026. Listicle test: does the engine include you in the list?
• [Your brand] reviews. Direct-name test: does the engine know what your product does?
• Alternatives to [your top competitor]. Adjacency test: does the engine surface you as a competitor's alternative?

The first one is the listicle test SEO tools have run for fifteen years; AI engines treat it surprisingly similarly.

Step 6. Read the verdict (under a minute)

You now have a small but real measurement. The next section is how to read it.

Coverage 0 of 4 engines

Invisible. This is the Invisible 10 territory. The B2B brands we see in this state are usually well-funded, have working SEO programs, and ship product regularly. They are simply not yet read by the AI engines their buyers are starting to ask.

The fix is rarely "write more content." Almost always it is two things: the technical floor (your homepage's crawlability for the AI bots), and the third-party citation footprint (whether other sites the engines have indexed reference your brand by name). The technical floor is fixable in days. The citation footprint takes months.

Coverage 1 to 2 of 4 engines

Partial. The pattern matters more than the count. Which engines cite you?

If Perplexity is one and ChatGPT or Claude is not, you are probably visible to live-web retrieval but not to training-data recall. Your brand exists on the indexed web; it is not yet baked into the model weights. Fix: keep doing what got you into the live-web index; double down on the third-party content the engine that does cite you read.

If ChatGPT or Claude cites you and Perplexity does not, you are in training data but not surfacing in live-web retrieval. Often that means your brand is mentioned in older content but newer content has dropped you. Fix: refresh the third-party content cycle.

Coverage 3 to 4 of 4 engines

Well-cited. Rare for B2B brands as of mid-2026. Run the accuracy check before celebrating: are the engines describing your product correctly, or are they listing your competitors' offerings as yours, or naming an old version? Accuracy errors at high visibility are a bigger win to fix than the next citation increment.

What this score is not

• Statistically rigorous. One prompt per engine is N=1; a proper audit uses N=3 to surface confidence intervals.
• Stable over time. Engine responses drift, sometimes weekly.
• Comparable across regions. AI engines often serve different content to US, EU, and APAC users.
• Free of personalization. If you ran the audit logged in, your account history influences the answers.

The score is directional. If a ten-minute audit says you are invisible, a proper measurement says the same thing with more decimal places.

If your coverage is 0 of 4: start with the technical floor

The AI crawler checklist covers the homepage-level checks before any content strategy. Are GPTBot, ClaudeBot, PerplexityBot, and Google-Extended allowed by your robots.txt and not silently blocked by your WAF? Does your homepage's core product positioning render server-side, or only after JavaScript hydrates? Those are the cheap fixes. Configuration, not content.

If your coverage is 1 to 2 of 4: find the citation pattern

Search Google and Bing for "your brand" "your category" and look at the top ten organic results. That set of pages is roughly the corpus the AI engines have read about your brand. The engines that cite you are reading some of those pages; the engines that miss you are reading different ones, or none.

The forensic question is which third-party content drove the citation. Often a single industry-publication article, a single G2 listicle, or a single podcast transcript moves the needle for one engine and is invisible to another. That single piece of content is the template for the next.

If your coverage is 3 to 4 of 4: check accuracy and competitor density

Re-read the twenty trials. For each citation of your brand, is the engine describing the product accurately, or is it confusing you with someone else? For the prompts where a competitor was named instead of you, what did the engine know about that competitor that it did not know about you? The answer is rarely "the competitor has more content." It is often "the competitor is named in one specific third-party context the engine respects."

If you want a deeper measurement without doing it yourself

Web Cited's tools come in a free tier and a paid tier. The Free Snapshot at /snapshot is a single-prompt citation check on your domain plus a homepage fix-list with the technical-floor checks that drive citation share; free in exchange for your email. The full SXO Audit ($5,000) runs the deep version: twenty-five buyer prompts across six engines (ChatGPT, Claude, Gemini, Perplexity, Grok, DeepSeek) with N=3 multi-trial sampling, plus the technical-floor checks across your full site, plus the third-party citation map, often delivered within hours.

The point of being measurable is not paywalling the measurement. If you can run a ten-minute audit on your own laptop and reach the same yes-or-no verdict our Free Snapshot would give you, run the ten-minute audit. We publish the manual version because we would rather you know whether you are invisible than buy a tool to find out.

What the paid audits add is the parts a manual audit cannot do in ten minutes: multi-trial sampling so the result is reproducible, six engines instead of four, twenty-five prompts instead of five, the technical-floor checks, the third-party citation map, and a Playbook your engineers can ship from in their next sprint. Those things are what the $5,000 covers. The yes-or-no screening question is what the manual audit covers.

The audit takes about ten minutes. The hardest step is being honest about the five buyer prompts. Most teams overestimate how often their brand appears for prompts they have never actually run.

Try the Free Snapshot   Order the SXO Audit ($5,000)

Or run the manual version above. The audit is free either way. The difference between manual and paid is how many decimal places you want on the answer.