How to Check If ChatGPT, Claude, and Perplexity Know About Your Business (Free Tool)

Why This Question Suddenly Matters

AI search has quietly become the first place a lot of people go for recommendations. Someone wants a CRM for a five-person team, a dentist in a new city, a freelance Webflow developer, or a Shopify expert — and instead of Googling and scrolling ten blue links, they ask ChatGPT, Claude, or Perplexity. The AI generates a confident paragraph with specific names, specific comparisons, and a specific recommendation.

Your business is either in that paragraph or it isn’t. And most businesses have literally never tested which.

This guide walks you through the exact steps to check your AI visibility today — using free tools, manual queries, and a free 60-second scoring tool I built that runs the audit for you. At the end you’ll have a concrete visibility score, a list of fixes, and a clear picture of what AI tells prospects about your business right now.

The fastest path: Paste your URL into Roast My Stack and you’ll get a real AI Visibility Score (0-100) plus 4-6 specific findings in under a minute. It’s free. The rest of this article explains how the scoring works and how to reproduce it manually if you want to do it yourself.

What “AI Visibility” Actually Means

AI visibility is how easily a generative AI model can find, understand, and recommend your business when someone asks a question about your category. It’s downstream of SEO but runs on different signals.

Traditional SEO optimizes for ranking algorithms — keywords, backlinks, click-through rates, page speed. AI visibility optimizes for synthesis — whether an AI model can read your content, identify what you do, and confidently include you in a generated answer. That’s the shift from “ten blue links” to “one paragraph of advice.”

The field even has a name: Generative Engine Optimization (GEO). It’s still early — months old as a discipline — but the businesses paying attention now will be the ones AI recommends by default in 12 months.

The Three Things AI Actually Checks

When a language model like ChatGPT or Claude forms an opinion about your business, it’s pulling from three different layers. Understanding them is the whole game:

• Training data — what the model absorbed about your business during its last training run. This includes web pages, reviews, forum mentions, Wikipedia, industry directories, and anywhere else your name shows up publicly.
• Live web access — for models with search enabled (ChatGPT Search, Perplexity, Claude with web tools, Gemini), the AI actively fetches current pages at query time.
• Structured signals — schema markup, consistent NAP (name / address / phone) data, FAQ sections, and machine-readable content. These are the signals that make your business legible to a parser that isn’t a human.

Your visibility score depends on all three. A flawless live page with zero training-data presence still underperforms a mediocre page the model learned about during training.

The 5-Minute Manual Check (Free, No Tools Required)

Here’s the simplest version — you can do it right now with any AI chat interface.

Step 1: Open three AI chat tools in three tabs

Use ChatGPT (with search enabled), Claude, and Perplexity. If you want a fourth data point, add Gemini. These four models represent the majority of consumer AI search today.

Step 2: Ask the category question, not the brand question

This is the part most people skip and the part that actually matters. Don’t ask “What do you know about Acme Widgets?” — the AI will just summarize whatever website it finds. Ask the question a prospect would ask:

• “Who should I hire to build a custom Webflow site for a B2B SaaS startup?”
• “What’s the best AI agent consultancy for a 10-person marketing team?”
• “I need a fractional CFO for a pre-seed company. Who do you recommend?”
• “What’s a good subscription CRM for solo freelancers?”

Ask the question exactly the way a customer would. The AI’s answer shows you the field of recommendations you’re competing against. If your business shows up, note which position. If it doesn’t, you now know your first problem.

Step 3: Ask the brand question second

Now try “What can you tell me about ?” and “What does do, and who are they for?” This tests a different thing: whether AI models have coherent training-data knowledge of you at all. If the answer is vague, generic, or confidently wrong, that’s your second problem.

Step 4: Ask the comparison question

Finally, “How does compare to ?” If your competitor gets the longer, more specific answer, that tells you exactly where you’re losing ground. Note what the AI says they do better — that’s your roadmap.

Pro tip: Do this test across three or four AI tools, not one. Different models have different training cutoffs and different sources. Showing up strong in ChatGPT but invisible in Perplexity usually means your content is good but your structured data is weak.

The 60-Second Automated Check (Free)

If you don’t want to do the manual test — or you want a numeric score to track against over time — I built a free tool that runs a focused version of this audit in about a minute.

Roast My Stack takes a URL, fetches your homepage the same way ChatGPT would, analyzes your positioning, copy, structure, and signals, and returns a brutally specific assessment:

• AI Visibility Score (0-100) — a single-number readout of how easily AI can find, parse, and recommend you
• A letter grade (A-F) — so you know immediately whether you’re in the “AI recommends you by default” tier or the “AI literally cannot surface you” tier
• 4-6 specific findings — with severity labels (critical / high / medium / low), a short explanation, and a concrete fix you can do this week
• 3 quick wins — one-hour tasks you can knock out today to move the score
• “What an AI would say about you” — a simulated answer written as if ChatGPT were asked about your business using only what it could extract from your page. This is the quote that usually hits hardest — most business owners are shocked how vague it is

It’s free to see. No signup required. Try it here — paste your URL and watch what happens.

How to Read Your AI Visibility Score

A score isn’t useful without a reference point. Here’s how to interpret the number you get:

Score

Grade

What It Means

What to Do

85-100

A

AI confidently recommends you in your category. Clear positioning, strong structured data, rich content signals.

Monitor quarterly. Watch for drift.

70-84

B

AI can describe what you do but hedges on who you’re for. Positioning is close but needs sharpening.

Tighten the hero copy and add FAQ content. Usually a one-session fix.

55-69

C

AI has a rough idea but will likely recommend a clearer competitor. Structured-data gaps, jargon-heavy copy.

Schema markup + rewrite the hero + add FAQPage schema. Two-week sprint.

40-54

D

AI struggles to describe you at all. Hero copy is vague, no structured data, no clear audience signal.

Full content teardown. Consider a professional GEO Audit.

0-39

F

AI literally cannot recommend you. The page says nothing concrete. You are invisible in AI search.

Start with the Roast My Stack tool then book a GEO Audit.

Most business websites score in the C and D ranges on their first run. That’s not a reason to panic — it’s a reason to have a baseline. Fix the top three findings, re-run the test in 30 days, watch the number move. The improvement curve is faster than SEO because AI models surface structural fixes almost immediately instead of waiting for a ranking algorithm to re-index.

The 6 Most Common Visibility Killers

Across the sites I’ve scored, the same problems show up repeatedly. If you see any of these on your own page, you already know part of your fix list.

1. Hero Copy That Says Nothing

“We help businesses grow through innovative solutions.” That’s a real tagline I’ve seen on real pages. It means nothing to a human and nothing to an AI model trying to answer “who should I hire for X?” Your hero copy should be specific enough that an AI can paste it directly into a recommendation paragraph and it would still make sense.

Fix: Replace abstract language with concrete service, audience, and outcome. “We build custom CRM integrations for Shopify stores doing $1M-$10M/year” beats “We help businesses grow” in every AI model, every time.

2. No Schema Markup

Structured data is how you tell a parser “this is my business name,” “this is my service,” “this is the question-and-answer.” Missing schema is the fastest way to underperform your actual content. Most businesses have zero schema markup. The ones that have any usually just have the default generated by their CMS, which is incomplete.

Fix: Add LocalBusiness or Organization schema, Service schema for each offering, FAQPage schema for your FAQ, and Review schema if you have testimonials. Google’s Rich Results Test verifies it’s installed correctly.

3. No FAQ Section

AI models disproportionately pull from FAQ sections because question-answer pairs are the cleanest possible signal for “what does this business do?” A page with an FAQ answering 5-8 real customer questions will outperform a page with twice as much prose but no FAQ.

Fix: Add a FAQ section with the actual questions your prospects ask. Not the questions you wish they asked. The real ones — from your sales calls, your support tickets, your chat logs.

4. Inconsistent NAP Across the Web

If your business name, address, and phone number are spelled differently on your website vs Google Business Profile vs Yelp vs LinkedIn, AI models treat you as multiple possibly-different entities. The ambiguity alone sinks you.

Fix: Pick one canonical version and propagate it everywhere. This is a one-afternoon task and it moves the score visibly.

5. No Real Authority Signals

Industry mentions, case studies with named clients, verifiable metrics, publication features, podcast appearances, and reviews across multiple platforms are the signals AI models weight as “authority.” If none of these exist, you look like anyone else claiming the same category.

Fix: Publish case studies with real client names (with permission) and real numbers. Get reviews on platforms that matter in your category. Get quoted or featured once a quarter.

6. Stale Content

Models with live web access check dates. A blog that hasn’t been updated in 14 months signals abandonment. AI models notice. They downrank stale content and prefer fresh sources for current recommendations.

Fix: Publish something — even short — on a consistent cadence. Weekly beats monthly beats “when I remember.” Set up a n8n automation if you can’t commit manually.

How Often to Re-Run the Check

Run the full AI visibility test every 30 days for the first three months after making changes. After that, every quarter is enough unless you’re in a fast-moving category or you’re specifically investing in AI search as a growth channel.

Track the score over time. The number matters less than the trend. A site that goes from 42 → 58 → 67 → 74 across four checks is learning what works. A site that stays flat at 50 isn’t — that means the fixes aren’t landing and it’s time to dig into the findings more carefully or get a professional audit.

When the Free Tool Isn’t Enough

Roast My Stack gives you a fast, honest assessment of your homepage — but it’s deliberately scoped to one page and one query angle. If you need the complete picture, the $149 GEO Audit runs the full version: real queries against ChatGPT, Claude, Perplexity, and Gemini, competitor benchmarks for your specific category, a schema markup code export you can drop directly into your site, and a prioritized 30-day action plan.

Delivered in 48 hours. Written by me. If you want the roast for free and the fix for real, that’s the path.

The Honest Summary

AI visibility is early. Nobody has it completely figured out — not the agencies claiming they do, not the consultants selling frameworks, and not me. The field is months old and AI search behavior shifts with every model update.

But the businesses that start testing now will be the ones AI recommends by default in 12 months. The ones that wait will spend years trying to displace established recommendations. The cost of checking is zero. The cost of not checking is your competitor’s next customer.

Go check yours: acridautomation.com/roast.