How do I check AI visibility for a US small business website?

You check AI visibility by running a working set of buyer questions through the major AI chat engines, ChatGPT, Claude, Gemini, Perplexity, Grok, and DeepSeek, plus a quick look at Google AI Overviews. Then you score each answer. Note whether your business shows up, whether the facts are right, and whether the reader gets a clear next step. The work is simple. It takes about two hours the first time. You walk away with a baseline you can re-test in 30 days.

If you have asked ChatGPT about your own industry and watched it skip your business, you are not alone. Most US small businesses were built for Google in 2018. They were not built for AI search in 2026. The good news is the check is easy. The fix takes longer.

Why this matters now

US small businesses are buying AI tools faster than they are being found by them. The US Census Bureau's Business Trends and Outlook Survey shows AI adoption sat at about 18 percent of US firms at the end of 2025, with more than 20 percent of firms expecting to use AI in the first half of 2026. The SBA Office of Advocacy reports that small business AI use has roughly doubled in six months, from 6.3 percent to 8.8 percent, narrowing the gap with large firms.

That is the supply side. On the demand side, Americans use AI tools to find businesses, not just to write captions. Pew Research found about 18 percent of Google searches in March 2025 produced an AI summary. When an AI Overview shows up, click-through to the regular blue links roughly halves, from 15 percent to 8 percent. Pew also reports 34 percent of US adults have used ChatGPT, and as of early 2026, 36 percent had used it in the past week.

Here is the part that stings. Roughly one in four US small businesses still has no website at all, and many that do have a thin one. If a business is invisible to plain Google search, it is invisible to AI search too. The AI engines pull from the same web Google reads.

You do not need to fix every problem this week. You just need to know where you stand.

What "AI visibility" actually means

AI visibility is the share of buyer questions where an AI engine names your business, links to it, or cites it as a source. It is close to Google ranking, but it is not the same.

Three things make it different.

First, AI engines summarize rather than list. A Google search shows ten links. A ChatGPT answer often names two or three businesses and skips the rest. The cut-off is sharper.

Second, the same question asked twice can give different answers. AI engines are probabilistic, so brand mention rates can swing widely across runs of the same prompt. You need more than one sample to trust the result.

Third, structured data matters more. Princeton's peer-reviewed GEO paper (Aggarwal et al., KDD 2024) found that adding clear stats and citations to a page lifted the chance an AI engine would quote it by 37 percent. Schema markup, named expertise, and clear numbers are signals AI engines weight heavily now.

The four-step check

Use this order the first time. It takes about two hours.

1. Pick a prompt set that mirrors how a real buyer would ask.
2. Run each prompt on each engine, more than once, in fresh sessions.
3. Score what you find across five simple lenses.
4. Find the gaps an AI engine would need closed before it could name you.

Each step is below.

Step 1: Pick your prompt set

Aim for around 10 prompts. Fewer and you are guessing from noise. Many more and you are paying for repetition. Mix three intent levels.

• Top of funnel. Broad questions a stranger might ask. "Who are the best HVAC contractors in Phoenix?" "What dental office should I take my kid to in Brooklyn?"
• Mid-funnel. Compare and shortlist questions. "Compare three local accounting firms in Austin for an LLC owner." "Which marketing agencies in Nashville work with restaurants?"
• Bottom of funnel. Direct brand checks. "Tell me about [your business name]." "What does [your business name] do and where are they based?"

Write the prompts in plain words. Write them the way your customer would type them, not the way you would. If you sell a service, name the city, the neighborhood, or the metro area. AI engines weight local context heavily. ZIP code prompts work for some categories, but most buyers ask by city or neighborhood first.

If I were you, I would pick three or four prompts in each intent band. Save them in a notes app or a sheet so you can re-run the same set in 30 days. Keep the set roughly the same size each time. Only the content of the prompts changes when your services change.

A solo dental practice in Denver ran her first scan with about a dozen prompts across the six engines. She showed up on Perplexity four times, on ChatGPT once, and on the other four engines not at all. Her gut said she was "doing fine online." The data said the opposite. That gap, between gut feel and a real baseline, is the whole point of the exercise.

Step 2: Run each prompt on each engine

Use six chat engines. Together they cover the bulk of AI search behaviour. They act differently and pull from different data, so a business can be named on one and missed on the others.

• ChatGPT (free or Plus). Sends the largest share of AI chatbot traffic to websites in 2026, per Pew tracking.
• Claude (claude.ai). Anthropic's chat product. Strong on research, analysis, and policy-shaped questions.
• Google Gemini (gemini.google.com). Google's standalone chat product, not the same as AI Overviews.
• Perplexity. Closer to a search-and-cite model. Always shows sources.
• Grok (x.com). xAI's chat product, growing share inside the X platform.
• DeepSeek. Open-weights model with a free chat interface, growing share in technical and B2B queries.

Around 10 prompts across these six engines gives you enough samples to see patterns clearly. That is the working baseline.

Two related surfaces are worth a separate quick check:

• Google AI Overviews, the AI summary that appears inside Google search results.
• Microsoft Copilot, the chat layer integrated into Bing and the Microsoft 365 ecosystem.

For each prompt, on each engine, do this:

1. Open a fresh session or an incognito tab. AI engines remember context.
2. Ask the prompt as written. Do not soften it.
3. Run the same prompt at least twice. Variation matters.
4. Save the answer. A screenshot is fine.
5. Note whether your business is named, linked, or cited as a source.

Most owners are surprised by how often the answer is "no." Take a breath. The point of the check is to know.

Step 3: Score what you find

Five lenses cover the ground. Tally each prompt against each one.

• Visibility. Are you named at all? Yes or no.
• Favorability. When named, is the tone positive, neutral, or negative?
• Factuality. Are the basics right? Address, services, hours, founder name.
• Actionability. Did the engine give the reader a clear next step (a link, a phone number, a citation)?
• Consistency. Does the same prompt give a similar answer across runs and across engines?

A simple grid works. Prompt down the side, engine across the top, score in each cell. Color the misses red so the pattern jumps out.

You are not chasing perfection. You are looking for patterns. A common one is "named on Perplexity, missing on ChatGPT." That tells you the issue is likely weak citations, not weak basics. Another is "named on ChatGPT, but with the wrong address." That is a factuality problem driven by stale third-party listings.

Step 4: Identify the gaps

Three gaps cover most cases.

Gap one: nothing for the AI to find. This is the roughly one-in-four small business stat. No website, or a site with thin pages. The fix is foundational. A clear Home, About, Services, and Contact page, written as real text rather than baked into images, is the minimum.

Gap two: a website exists but the structured data is missing. Open one of your pages, view the source, and search for application/ld+json. If you find nothing, you have no schema. Google's own Search Central docs name Organization schema as the baseline for any business. They name LocalBusiness schema for any business with a storefront or service area. AI engines lean on this markup hard for entity matching. A small business site without LocalBusiness schema is, in effect, telling AI engines almost nothing about itself in machine-readable form.

Gap three: few or no citations from sources AI engines trust. AI engines pull from real media, peer-reviewed research, official directories, and well-cited articles. If your business is not named in any of those, you have what GEO researchers call a citation desert. The fix is slow but it lasts. Get listed in real industry directories. Write one or two original pieces for outlets your customers read. Make sure your About page names a real person with real credentials.

What good AI visibility looks like for a US small business

There is no official benchmark yet. Here is a working one based on what shows up in healthy first scans.

• Named in 60 percent or more of bottom-of-funnel prompts. If "tell me about [business name]" cannot find you, the brand has no AI footprint.
• Named in 20 to 40 percent of mid-funnel prompts. Compare questions are tough. A spot in this range means you are in the consideration set.
• Named in 5 to 15 percent of top-of-funnel prompts. This is the hardest tier. A small business should not expect to win broad queries.
• Factuality at or near 100 percent when you are named. A wrong address or a wrong service line is worse than a miss.

Re-test in 30 days, then again at 90. A baseline you cannot compare to is just a snapshot.

Tools and resources

You can do the whole check by hand with a notes app and the six engines above. That is the cheapest option and it teaches you the most.

If you want to automate the prompt runs and scoring, a few products track AI visibility for a monthly fee. Tools in this space include HubSpot's AEO Grader, Otterly, Peec AI, and Get Recommended. Choose by which engines and signals you want covered, and by which region the tool understands. Prices vary and the space is moving fast.

For the schema side, Google's Rich Results Test at search.google.com/test/rich-results checks Organization and LocalBusiness schema for free. Schema.org's own docs are the source of truth for which fields the major engines accept.

For the citation side, the work is human. Local trade journals, Chamber of Commerce listings, regional industry bodies, named-author guest posts on respected outlets, and Better Business Bureau profiles carry weight. AI-written content marketing blogs do not.

A few quirks worth knowing if you are in the US

• EIN and DBA mix-ups are common. AI engines often blur two entities that share a similar name. Make your legal name and any DBAs (doing business as) clear on your About page and in your schema.
• Local intent is granular. Americans search by city, neighborhood, and sometimes ZIP code. Local pages should name the neighborhoods or service areas you cover, not just the metro.
• Trade press is fragmented. A single Forbes or Inc. mention helps, but most categories are won by deep coverage in vertical trade outlets, not big national press.
• Regulated industries face a higher bar. In financial services, legal services, healthcare, or licensed trades, the factuality lens matters most. Wrong info about a regulated service can be a compliance issue, not just a marketing one. State-level licensing and disclosure rules often apply.

Pulling it together

The honest version is short. Pick a working set of around 10 prompts. Run them through the major chat engines. Score what you see. Write down your three biggest gaps. Pick one you can close this month. That is the whole method.

The first scan will tell you more than any marketing report you have read in the last year. It will likely be uncomfortable. That is fine. Discomfort is the start of clarity.

Run the check. Write down your three biggest gaps. Pick one to fix this month. Re-test in 30 days and look for movement, not perfection.