An AI visibility audit you can run on your own site this week
Ten checks, no tooling budget required. Most brands fail three of them, and two of those are one-line fixes.
Published 11 September 2026 · 8 min read · Kaada Nordic MarTech
Key takeaways
- Start by asking the assistants directly — their answers are your baseline.
- Blocked crawlers and JavaScript-only content are the two most common hard blockers.
- A plain facts page is the cheapest single improvement most sites can make.
- Third-party corroboration is what turns eligibility into citations.
You do not need a subscription to anything to find out whether AI assistants can see your business. The following checks take an afternoon and surface the blockers that matter, in the order they matter.
1. Establish a baseline
Ask Copilot, Perplexity and ChatGPT the questions your buyers ask: "best [category] for [audience]", "alternatives to [competitor]", "who offers [specific capability] in [market]". Record which brands are named and which sources are cited. Those cited pages are your real competition for the answer slot.
2. Check you are in the index at all
Search your exact brand name plus a product term in Bing, not only Google. Copilot's answers are grounded in the Bing index; plenty of sites rank respectably in Google while being thinly indexed in Bing.
3. Read your robots.txt
Look for rules affecting the assistant crawlers — among them BingBot, PerplexityBot, OAI-SearchBot, ChatGPT-User and ClaudeBot. Blanket blocks are often inherited from a template or added during a scraping scare, and they silently remove you from those assistants' answers.
4. View your page as a crawler does
Disable JavaScript in your browser and reload your key pages. Whatever remains is roughly what a crawler reliably reads. If your product description, pricing or location disappears, so does your ability to be quoted on it.
5. Test the one-sentence answer
For each important page, find the single sentence a model would lift to answer "what is this and who is it for?". If no such sentence exists, write one and put it near the top.
6. Publish a plain facts page
- What you sell, in category terms buyers use
- Price or pricing model, and what it includes
- Who it is for and who it is not for
- Markets and languages served
- Contact details and legal entity
This page rarely wins design awards and frequently wins citations, because it answers factual questions in one place without marketing hedging.
7. Align structured data with the visible page
Structured data should describe what a human sees. Prices, availability and organisation details that contradict the page body undermine trust rather than building it, and can trigger manual actions in classic search.
8. Date and source your claims
Any statistic worth citing is worth attributing. "Roughly 1.56 billion monthly visits (Semrush, July 2026)" is quotable. "Billions of users" is not, and a cautious model will pass over it in favour of a source that committed to a number.
9. Look for third-party corroboration
Search for your brand outside your own domain. If every result is a page you control, assistants have nothing independent to lean on. One credible editorial article, directory entry with real detail, or industry listing changes that materially.
10. Re-run the baseline monthly
Answer engines are not stable. Re-ask the same questions each month, keep the transcripts, and watch which sources appear. Movement in the cited sources tells you more about your progress than any single ranking number.
If you would rather not run this yourself, our free AI visibility check covers the automated parts and returns the gaps we find, with no obligation.
Sources
- Semrush domain overviews — msn.com, bing.com — Traffic and authority figures cited across this site.
- Published crawler documentation — Bing, Perplexity, OpenAI and Anthropic each document their user agents.
Get your brand into the answers
We publish an editorial article about your brand on MSN.com and hand you an AI SEO audit, brand DNA report and competitor snapshot. Ten pilot seats, €1,500 per month.
