How to check whether AI recommends your business: a test protocol for ChatGPT, Gemini and AI Overviews

What exactly we're checking and why it matters

Customers increasingly ask questions directly instead of typing keywords. AI Overviews in Google show an answer at the top instead of a list of links. ChatGPT and Gemini also let users search for information. In each of these places, your business can be mentioned or not.

The three surfaces you need to test:

• AI Overviews and AI Mode in Google — appear above traditional results.
• ChatGPT search — works on OAI-SearchBot.
• Gemini — Google's assistant; the Google-Extended token in robots.txt decides whether site content can be used for training Gemini models.

Google Search Central explains that SEO rules are the same as for regular search — no special requirements for AI. OpenAI uses OAI-SearchBot for search and GPTBot for training — you can control them separately. Google-Extended only affects Gemini, not Google search results.

Test protocol — 15-25 queries in four buckets

Conducting a proper test requires preparation. Entering a single query once isn't enough — AI answers vary depending on phrasing, location, and timing. Here's a ready-made set of queries divided into four categories.

Bucket 1: Service + district or city

These queries simulate a customer looking for a specific service in the area. Examples:

• "dentist Mokotów Warsaw"
• "car mechanic Ursynów"
• "hair salon city center Kraków"
• "air conditioning service Wrocław"

Record whether your business was mentioned, and if so, at what position in the list and what sources AI showed.

Bucket 2: Customer's problem in their own words

Here the customer doesn't yet know the service name, just describes the problem. Examples:

• "tooth hurts in the evening where to go"
• "car won't start in the morning"
• "need a haircut for tomorrow"
• "air conditioning works but doesn't cool"

These queries show whether AI understands intent and can recommend a business based on offerings, not just keywords.

Bucket 3: Comparison "X or Y"

Customers often compare options before deciding. Examples:

• "better dentist Mokotów or Ursynów"
• "computer repair or replace"
• "expensive or cheap hair salon in city center"

Here you check whether AI mentions specific businesses in a comparative context.

Bucket 4: Your business name

Direct queries about your business name:

• "[your business name] reviews"
• "[your business name] opening hours"
• "[your business name] contact"

This is the baseline test — whether AI "knows" your business exists at all.

Results sheet — table for entering data

Run each query three times at least an hour apart. AI answers differ between attempts, so a single result isn't reliable.

How to measure honestly — principles of reliable testing

For test results to have value, you must eliminate variables that could affect AI's answer. Follow these rules:

Clean session without logging in. A logged-in user sees personalized results based on search history and location. Log out or use incognito mode. In Chrome, Ctrl+Shift+N opens an incognito window.

Same location. AI Overviews in Google show results dependent on the user's location. If you test from Warsaw but potential clients are in Poznań, the results will differ. Use the same location for every test or note which city you're measuring from.

Same time of day. Answers may differ depending on time of day due to index updates and model variability. Take all measurements at similar times or distribute them systematically.

Three repetitions of the same query. Record results of each attempt separately. If the business appeared once out of three, that's also information — about visibility instability.

Record sources. If the answer shows sources, record the exact links — this will show you what AI "knows" about your business and where it gets the data.

Who can read your page — robots.txt

Before making improvements, check who can even read your page. The robots.txt file is the first line of defense — it specifies which crawlers may visit (crawl) your site and which may not.

How to check your robots.txt in a minute

1. Enter in your browser address: yourdomain.com/robots.txt
2. Look at the User-agent and Disallow sections
3. Look for mentions of: OAI-SearchBot, GPTBot, Google-Extended, Googlebot

If you see something like:

User-agent: OAI-SearchBot
Disallow: /

Your site won't appear in ChatGPT search but Googlebot can still read it — Google Search works independently. That's not an error — it's a conscious choice, but you need to know about it.

If you don't have a robots.txt file, that's also information — by default, all crawlers may crawl your site.

A separate issue is Google-Extended. If you see:

User-agent: Google-Extended
Disallow: /

This means you've blocked content use for Gemini and Vertex AI training, but search results in Google and AI Overviews aren't affected.

Google Business Profile and Apple Business — the visibility foundation

For local results, Google Business Profile is the primary source of business data. Google Business Profile Help confirms it's one of the main sources of local business information.

Three local visibility filters

Google's official help explains that local results depend mainly on three factors: relevance, distance, and prominence. Google Business Profile Help confirms these three filters. You can't change distance — your business is where it is. But you can influence the other two:

• Relevance — how well your Business Profile matches the search. Enter the full range of services, exact address, all categories describing your activity.
• Prominence — visibility online, number of reviews, responses to them, information currency.

Naming and hours rules

Google forbids adding keywords to the business name, but many owners do it hoping for better rankings. The effect can be the opposite — the profile can be suspended for violating rules, and then the business disappears from results entirely.

Opening hours must be current and match reality. If the profile shows summer hours but the business has already switched to winter, AI will give customers wrong information.

Apple Business — a second source

Apple Maps also displays business information. If you have an Apple Business Profile (available at apple.com/business), enter the same data manually — name, address, phone, hours, category should be identical in both places.

NAP data consistency — the rule

NAP stands for Name, Address, Phone. These three elements must be identical everywhere: on the website, in Business Profile, on industry portal profiles, in Apple Business, on business cards, and in every place where the business is listed. Discrepancies may affect the consistency of information that AI passes to users.

LocalBusiness structured data — what it gives and doesn't guarantee

Google's LocalBusiness documentation explains how to mark up business data in site code so algorithms can read it. LocalBusiness documentation describes required fields and JSON-LD examples. JSON-LD structured data lets you describe name, address, phone, opening hours, and other details in machine-readable format.

Key elements to implement

{
  "@context": "https://schema.org",
  "@type": "LocalBusiness",
  "name": "Business Name",
  "address": {
    "@type": "PostalAddress",
    "streetAddress": "Street and number",
    "addressLocality": "City",
    "postalCode": "postal-code",
    "addressCountry": "PL"
  },
  "telephone": "+48123456789",
  "openingHoursSpecification": [
    {
      "@type": "OpeningHoursSpecification",
      "dayOfWeek": ["Monday", "Tuesday", "Wednesday", "Thursday", "Friday"],
      "opens": "09:00",
      "closes": "17:00"
    }
  ]
}

openingHoursSpecification lets you precisely describe opening hours for each day of the week — including breaks and holidays.

What structured data gives

Good JSON-LD structure increases the chance that Google will display extended business information in search results: exact hours, location on map, phone number as a clickable link. This doesn't guarantee appearing in AI Overviews, but gives the algorithm material to work with.

What it doesn't guarantee

Structured data is a supporting tool, not a recipe for appearing in AI.

Content that AI can cite — what to write on your site

If AI is to recommend your business, it needs something to draw from. The website is the primary source. Google's content documentation recommends that content answer the questions: Who, How, and Why. Content should be created for people, not search engines.

Questions worth answering on your site

Who runs the business — not as a generic "we are an experienced team" but specifically: owner's name, experience, specialization. How the process works — from first contact through completion to wrap-up. Why choose this business — what sets it apart from competitors, what are the terms, guarantees, what's included in the price.

The more specific the information, the greater the chance AI will find an answer to the customer's question.

Our measurement — how many businesses lack their own content

This means they have no original content that AI could read and potentially recommend. Their only source of information is a profile on an external platform — where data is limited to the minimum.

If your business has only an external platform profile, consider setting up a simple site with basic information — even one page with services, contact details, and business description.

How to read the results — decision matrix

After conducting the test, you have data. Time to interpret it. Use a simple matrix:

What to do in each situation

30-day action plan without a contractor

Week 1: Measurement

Choose 15-20 queries from the four buckets. Run the test in AI Overviews, ChatGPT, and Gemini. Record results, compare with three competitors.

Week 2: Business Profile and robots.txt

Check Business Profile data completeness: hours, category, attributes. Make sure OAI-SearchBot isn't blocked.

Week 3: Website and structured data

Add LocalBusiness JSON-LD (or use SEO plugin). Fill in content answering Who, How, Why questions.

Week 4: Repeat measurement

Run the test again and compare with week 1 results.

What visibility looks like when the system manages it

When you manually test business visibility in AI answers once a month, you have a snapshot of the current state. A systematic approach lets you automate this process and react to changes in real time.

The GEO AI service in Aura consists of the following stages: brand presence audit in search and selected AI answers, entity and topic mapping, content and schema, implementation, and monitoring (continuous). At each stage we work on giving algorithms something to read about the business.

We don't promise a place in AI answers — it depends on algorithms. But regular monitoring helps detect changes and react.

If you want to see what such a report looks like and what data we collect — check out the GEO AI service. If you need a comprehensive approach to online visibility, start with SEO services. If you want to evaluate your current situation first, check your Google Business Profile. If you're interested in automated reporting, see AIREPORTS, and for basic data analysis.

For more insights on how other businesses handle visibility, see: The Restaurant Facts Assistants Read About You, Responding to Google Reviews, Reporting automation, and Why customers don't leave inquiries.

Frequently asked questions

Do I have to pay to be visible in AI?

No. There's no separate fee for appearing in AI Overviews or ChatGPT. The same SEO rules apply as for regular search.

Does blocking in robots.txt mean no one will find me?

Only partially. Blocking Googlebot in robots.txt means the crawler won't scan the page, but it doesn't protect against indexing. The page can be indexed without content (just the domain name) if it's linked from other places. To protect against indexing, you need to add noindex in meta tags or HTTP headers. OAI-SearchBot operates independently — if you block one crawler, the other can still read your site.

How long does manual visibility testing take?

Preparing the spreadsheet and running the first test takes about two hours. Subsequent tests — about an hour each if you use the same set of queries. With regular repetition (once a month), that's about an hour per month.

Do I need to know code to add structured data?

No. There are ready-made tools that generate JSON-LD code from a form. In WordPress, SEO plugins (Rank Math, Yoast) have this built-in. Just fill in the fields in the plugin panel.

What to do if AI mentions competitors but not my business?

First, compare data: check competitors' Business Profile — what categories, hours, attributes they have. Then analyze their website — what content they publish, how they answer Who, How, Why questions. Finally, fill in missing elements on your site and repeat the measurement after 2-3 weeks.

Do reviews affect AI visibility?

Indirectly, yes. Reviews affect "prominence" in Google's local algorithm, and that's one of the three filters (relevance, distance, prominence). Additionally, reviews contain keywords and phrases customers use — AI can read and use them in answers. Responding to reviews additionally increases the amount of content associated with the business.

How often should I repeat the test?

At least once a month to track changes. Ideally — before each major change on the site or in Business Profile. After adding new services, changing hours, adding new content — it's worth checking if AI has "noticed."