Getting a new brand found by AI assistants, not mistaken for another

We publish this one because it is ours. A new domain, a new company name, and an assistant that confidently answered questions about midalaxy.com with details belonging to Midaxo — a different business entirely. The obvious assumption was that something was broken.

Nothing was. DNS resolved correctly across four independent resolvers, TLS was valid, robots.txt allowed every crawler we cared about, the sitemap parsed, redirects were clean, and a different vendor’s fetcher retrieved the exact URL without complaint. Every technical check passed, which is the least useful result an investigation can return — until you take it seriously.

The real failure was one of identity. An assistant asked about an unfamiliar string does not fetch it; it searches, and search engines are built to be forgiving about spelling. With nothing indexed anywhere containing "Midalaxy", the closest confident match won, and that match was a company with a decade of citations behind it. We were not being blocked. We were being autocorrected.

That reframed the work from debugging to establishing an entity. Structured data gained a stable @id, a named founder, an explicit topic list and sameAs links to profiles that actually existed — so the name had something to anchor to. Canonical hosts were consolidated onto permanent redirects rather than temporary ones, which is the difference between accumulating ranking signals and repeatedly discarding them.

Distribution mattered more than markup. We wired IndexNow to Bing’s endpoint directly, pushing all 480 URLs the moment they change instead of waiting on an organic crawl — worth doing precisely because ChatGPT’s search reads Bing’s index, so a page Bing has not seen is a page the assistant cannot find. The generic aggregator endpoint proved unreliable from our network; Bing’s own accepted the full set first try.

Twelve days after the first check returned nothing at all, Search Console reported 85 pages carrying impressions, 33 of them averaging a first-page position. The measurement discipline mattered as much as the fixes: we read totals from the Pages report rather than Queries, because Queries withholds low-click rows and had us briefly reporting zero clicks when there were two.

The honest ending is that the original problem is not solved. Brand searches for "Midalaxy" are still zero, which is the number that decides whether an assistant stops guessing — and it is the one we now watch, rather than the one we would prefer to report.