Citation Audit for ChatGPT, Perplexity and Claude Search

In about a third of the citation audits we have run, the reason a brand was absent from ChatGPT and Perplexity had nothing to do with its content. It was a 403. The CDN rule set returned Forbidden to OAI-SearchBot and PerplexityBot while waving Googlebot through, because the managed bot category treats them as scrapers by default and nobody had looked. That costs an afternoon to fix, and no volume of content work compensates for it, so it is the first thing we test on every engagement. Three engines, three retrieval paths: ChatGPT search leans on its own index and on Bing, so a domain missing from Bing is close to invisible there whatever its Google position. Perplexity crawls with its own agents and corroborates hard against forums, comparison sites and review platforms, which in Australian health and property means Reddit, Whirlpool, ProductReview.com.au and the portals rather than your blog. Claude cites a narrow set per answer and favours sources that state a fact plainly with a date attached to it. The same page is often retrievable by one and invisible to the other two, which is why a single blended visibility number would hide the only thing worth knowing. What we test: Robots.txt read line by line for GPTBot, OAI-SearchBot, ChatGPT-User, PerplexityBot, Perplexity-User, ClaudeBot, Claude-SearchBot, Claude-User, Google-Extended, Applebot-Extended and Bingbot, including the wildcard and ordering interactions people get wrong. Live fetches with each declared agent from outside Australia, comparing status code and returned HTML against a browser render, because these fetchers largely do not execute JavaScript and a page that assembles itself after hydration arrives as an empty shell. Cloudflare bot toggles, Vercel and Akamai rules, and rate limits that pass the first request and block the third. Bing index presence, and whether IndexNow is submitting anything at all. Only then the citation test: a prompt panel run against each engine, recording whether you are named, whether the mention carries a link, which of your URLs it points at, and which third-party domains carry your name in place of your own site. What you receive: Access findings written for your platform team, each with the offending rule, the expected status code afterwards, and a retest we run once you deploy. The full run log. A ranked list of the third-party domains that keep appearing in answers about your category, with what each of them currently says about you. What we will not do: If you are also blocking AI crawlers to keep your writing out of training corpora, say so at the start. You cannot be cited by a system you refuse to let read you. The vendors separate their training and search agents only partially, and that trade-off belongs to you. We will lay out what each agent does, and we will not quietly flip a setting or tell you that you can have both. We do not post to Reddit, seed threads or arrange reviews. We will tell you which threads decide your category and hand them over. For an AHPRA-regulated practice a solicited patient comment is regulatory exposure as well as a bad idea, and we will say no to it twice if needed. Who this is not for: Brands with no third-party footprint at all. If nothing outside your own domain mentions you, no reviews, no directories, no press, no forum threads, the audit reports that in one line and the work you need is not an audit. Also not for anyone who wants a single number for the board. The three engines disagree about you, that disagreement is the finding, and averaging it away would be the only thing in the report we could not defend.