Case-study-zero has the number because we ran it on ourselves first. Twenty buyer-intent questions, four AI engines, eighty scored answers. Never Behind’s name did not appear in a single one. Not low — zero. We’re publishing that before we’ve run a single ad, because softening it would be exactly the kind of thing this audit exists to catch.
Why did we audit ourselves before auditing anyone else?
Because a company that sells “find out whether AI engines cite you” can’t credibly ask a prospect to trust a method it hasn’t first pointed at itself. Before Never Behind approached a single client, we ran the same prompt set we use in every audit against our own name and let the engines answer honestly — no easier questions, no friendlier engine. The prompts were the exact buyer-intent questions a prospect shopping for this service would actually type: “best AI visibility audit service,” “who can help me get my business found in AI assistants,” “how do I get my company cited by ChatGPT.” If a method is going to tell a client something uncomfortable about their own visibility, it has to survive being pointed at us first. This page is that result, published as it came back.
What does an AI visibility audit actually measure?
Twenty buyer-intent prompts spanning three stages of the buying journey — problem (“why doesn’t my business show up when I ask ChatGPT for recommendations”), solution (“what does an AI visibility audit include”), and vendor (“best GEO consultant or agency for generative engine optimization”) — run against four engines: ChatGPT, Claude, Gemini, and Perplexity. That’s 20 × 4, eighty answers, each one read and scored on the same three-point scale we use for every client: cited (named with a source link), mentioned (named, no link), or absent. No cherry-picked prompts and no re-running a cell until the answer looked better. Every cell scored once, and kept.
What was the result?
Zero. Eighty of eighty cells came back the same way — Never Behind’s name absent, not mentioned, not cited. That held across all four engines and all three funnel stages: no engine we did better on, no stage — not even bottom-of-funnel prompts like “who can help me get my business found in AI assistants” — where we picked up partial credit. In those same 80 answers, a competitor was named outright in 59 of them, 74%. The demand behind these prompts isn’t hypothetical. Real buyers asking these exact questions get real vendor names back today. We just weren’t one of them.
What do 1,264 sources across 547 domains tell us?
Across the 80 answers, the engines cited outside sources 1,264 times, spread across 547 distinct domains. We checked: Never Behind appears on none of them. That’s the finding underneath the finding. If this were a ranking problem, better on-page SEO would fix it — it isn’t. The single most-cited domain in the whole matrix was a leading GEO tool’s own website, cited 53 times, almost entirely from its own guides and comparison pages rather than other people writing about it. Reddit was cited in 22 separate answers, the widest spread of any domain in the data, ahead of any single competitor’s own site. Search Engine Land was cited by all four engines. We showed up in none of the places these engines already trust and quote — which means the fix isn’t a better homepage. It’s a citation-graph problem: we’re absent from the sources the engines pull from, not badly ranked within them.
What did we do about it?
We ranked the fix list by expected impact on citation share, not by what was easiest to ship first. Publishing something on our own domain worth citing came first — a methodology page, a glossary, worked examples like this one — because none of the citation-graph work below it pays off if there’s nothing on neverbehind.com yet worth linking to. After that: pursuing coverage on the trade-press domain all four engines already cite, showing up genuinely in the Reddit threads where these questions get asked, and getting into the comparison content that already ranks GEO tools against each other. None of that is “improve our SEO.” It’s the same diagnosis, run on us, that produced the fix list above — treat the citation graph as the target, not the ranking.