AI Visibility Scans
Shows a business how AI search sees it, and what to fix.
With the rise of AI, the way people search is quickly moving towards AI overviews rather than simple Google queries. This has created an entirely new form of product discovery and a new funnel for businesses to optimize.
AI models write their answers from what they find online. When a question requires current information, the model runs a web search and builds its answer from the few pages it reads, and those pages decide which businesses get named. Because of this, making sure the sources AI uses for its answers promote your business is the best way to increase how often you are included in the answer.
How the Scan Works
The scan starts with a set of prompts a customer might ask, and where the company wants to come up. Claude writes the prompt set, and a person then edits and approves the list before anything runs. From there, a script is used to send each prompt to the model's API and record the answers.
As every AI model is built differently, it's important to check multiple models, and this is done through their APIs (to make sure there is no memory or context being used to bias the findings). Additionally, as models don't repeat their answers, it's important to ask the same model multiple times in order to improve the accuracy of the findings.
Every answer comes back with the sources it used, which is tracked across all prompts and saved for further analysis. The stack is a mix of scripts, API calls, and AI queries, with token usage tracked and minimized to reduce cost where possible without impacting quality.
A large set of prompts about the business: general industry questions, buyer-intent questions, questions about its reputation, and more.
Each prompt is sent to each AI model through its API, and the answers are recorded to be analyzed later.
The answers come back with their sources, so we know where AI is getting its info.
A script evaluates each prompt and its answers to determine where a business is winning, where a competitor is winning, and where nobody is winning.
A large set of prompts about the business: general industry questions, buyer-intent questions, questions about its reputation, and more.
- ChatGPT
- Perplexity
- Gemini
- Claude
- Google AI Overviews
Each prompt is sent to each AI model through its API, and the answers are recorded to be analyzed later.
- Competitors
- Directories
- Articles and guides
- Forums
- The business's own site often missing
The answers come back with their sources, so we know where AI is getting its info.
- Won its own site is a source
- Contested only a competitor's is
- Open neither is a source
A script evaluates each prompt and its answers to determine where a business is winning, where a competitor is winning, and where nobody is winning.
What the Scan Produces
Every answer is saved along with the websites the AI used as its sources. A script then gives each prompt one of three results: won if the business's own website is one of those sources, contested if a competitor's website is but the business's isn't, and open if neither is. Put together, it shows where the business comes up and where it doesn't.
It also shows which websites the AI models rely on most, sorted by type, like a directory listing or a ranked “best of” list. That tells the business which kinds of pages it needs to appear on.
Site Check
Alongside the scan, a script checks the business's own website for the basics that let AI read it: whether AI tools are allowed in, whether key details like the address and hours are labelled so machines can understand them, whether there's a list of all its pages for AI to follow, and whether the text shows up without extra code having to run first.
The most important part is checking which AI tools are allowed in. Each AI company sends its own automated visitors, called crawlers, to read websites. Every website can have a public rules file, called robots.txt, that turns any of them away. The script reads that file and checks the rule for nine AI crawlers. If a crawler is blocked, that AI can't read the site, so it can't recommend it or use it as a source.
- GPTBotOpenAIblocked
- ClaudeBotAnthropicblocked
- PerplexityBotPerplexitypartial
- OAI-SearchBotOpenAIallowed
- ChatGPT-UserOpenAIallowed
- Perplexity-UserPerplexityallowed
- Claude-UserAnthropicallowed
- Google-ExtendedGoogleallowed
- BingbotMicrosoftallowed
Critical2 AI crawlers blocked: GPTBot, ClaudeBot
WarningPerplexityBot can't read /blog/
Code reads the website's rules file and finds the rule for each crawler, or the general rule that covers all of them.
Blocked: kept out of the whole site. Partial: kept out of some pages.
If a crawler is blocked, that AI can't read the site, so it can't point people to it.
From Data to a Plan
Using all of the results from the visibility scan and the site check, an action plan is created that targets the most important things to do, like getting listed on the sources the models already cite or letting AI crawlers back in.
Each step comes from something the scan found. Open prompts, where AI isn't using any business as a source yet, are often the quickest wins. A page on the business's own website that answers that question directly can claim them. For contested prompts, the goal is to get the business onto the websites that already mention its competitors, like the directories and “best of” lists AI keeps going back to. If the site check found AI crawlers blocked or key details missing, those get fixed first, because nothing else works if AI can't read the site.
Every step lists the prompts it should win, and the steps are ranked by how much they should help against how much work they take, so the business knows what to do first.
If you're interested, reach out by email at michael.simunec02@gmail.com to request a scan and find out more about GEO for your business.