Getting cited by AI now matters as much as ranking in Google once did. When a buyer asks ChatGPT or Perplexity for the best option in a category, the name the assistant gives is the new first page. Everything under it is invisible.
So here is the uncomfortable number. Across the first 100 AI-visibility audits, 89% of these businesses were recommended by zero AI engines when a buyer asked for the best in their category. The engines usually recognized the brand. They named a competitor anyway.
The cause was rarely technical. Most of the sites were crawlable and fine. They just were not the answer.
Being known is not being recommended
Each audit asked six live answer engines — ChatGPT, Perplexity, Gemini, Claude, Grok, and Google AI Mode — two questions: do you recognize this brand, and do you name it when a buyer asks for the best in the category.
On recognition, the average business was known by about 3 of the 6 engines. On recommendation, it was named by almost none, and 89% by zero. Recognition is not the bottleneck. Being chosen is.
Finding 1: the engines are not interchangeable
The biggest surprise is that "optimize for AI" is the wrong frame, because the engines behave nothing alike.
Mapped across more than 20 engines by how easily a small brand earns a citation:
- Perplexity and Google AI Mode will cite a brand with no reputation behind it. This is where the fastest wins are.
- ChatGPT, Claude, and Grok rarely name a small business.
- Walled or language-locked engines (Baidu, Naver, Doubao) are hardest of all, since a brand effectively has to live inside their ecosystem to be cited.
Treating "AI" as one target wastes months. Start with the two doors that already open.
Finding 2: the demand is already there, unclaimed
Every audit surfaced content the business could rank for and get cited on: category search terms with real monthly demand and low competition.
Across the 100 audits, the reachable opportunity per site ran from about 460 to 348,000 searches a month, averaging around 62,000, with single keywords worth up to 265,000. Most of those terms carried a keyword difficulty under 10. Buyers are already typing these questions. The AI answer just points them at a competitor.
Finding 3: every audited site carried spam backlinks
Not most. All 100. On the median site, about two thirds of the backlink profile was spam, a median of 99 junk domains per site. AI engines and search both lean on trust signals, and a link profile that is mostly junk drags that trust down. The cheapest fix is a boring weekly habit: review new referring domains and disavow the junk.
The playbook the data points to
- Clean the backlinks, weekly. Assume the profile is dirty, because effectively every audited one was.
- Own the searches that already have demand. Publish clear, sourced pages answering the high-volume, low-difficulty questions.
- Get named in third-party lists and roundups. AI weighs consensus; one independent "best X" mention moves recommendation more than another owned-domain page.
- Structure every page for extraction. Put the direct answer in the first two sentences under a plain heading.
- Start with Perplexity and Google AI Mode. Win the doors that open before spending effort on the engines that stay shut.
The full report
All 100 audits, the full engine map, the score distribution, the unclaimed-search data, and the methodology are in the original report:
AI Visibility Audit: 100 Sites, 89% Invisible to AI →
To check which AI engines name a specific site today, there is a free audit — six engines, no signup — at fluenta.space/magnet.
Top comments (1)
curious if this works better for niche B2B sites or if it's mostly just for big consumer brands lol