If your business shows up when you ask Gemini or Perplexity but goes missing when you ask ChatGPT, the instinct is to blame the writing. In the case we document here, the writing was not the problem. ChatGPT was not reading the pages at all, because Bing had never crawled them.
This is worth knowing because the symptom looks exactly like a content problem, and treating it like one wastes months.
The Gap
We track a blog we operate for a local service business. Across fifteen questions, the mention rates split like this:
| Engine | Mention rate |
|---|---|
| Gemini | 79% |
| Perplexity | 67% |
| ChatGPT | 13% (2 of 15) |
Same pages. Same content. Same period. A gap that wide is not a quality signal — quality does not vary by engine. It pointed at plumbing.
We Got the Diagnosis Wrong Twice
We are writing down the wrong turns because they are the ones most people would take.
First wrong answer: "the blog isn't registered with Bing." It was. The webmaster tools showed the sitemap submitted on 8/9, successfully crawled on 8/12, and ten URLs discovered.
Second wrong answer: "it's just indexing lag, give it time." Closer, but too vague to act on. The coverage dashboard lumps everything together and hides the real state.
The precise answer only appeared in the URL Inspection screen, which reports three separate stages: discovered, crawled, indexed. The status read:
Discovered ✅ · Crawled ❌ · Indexed ⚪ — "Discovered but not crawled"
Indexed URLs: 0 of 10. Bing knew the addresses existed. It had never fetched the pages behind them.
It Was Not a Technical Block
Before concluding anything about authority, we ruled out the boring causes:
- Requesting the blog with a bingbot user agent returned HTTP 200 with the full HTML — no Cloudflare challenge, no 403, no 503
- robots.txt allowed everything, with no crawl-delay
- No noindex on article pages; canonical tags only
Worth noting: Bing's interface also displayed a generic warning about guideline compliance. That message is boilerplate shown on any uncrawled new URL, not a report of an actual violation. It sent us down one of the wrong turns above.
The Real Bottleneck Was Crawl Budget
The blog was a brand-new subdomain with zero inbound links. Search engines allocate crawl attention by priority, and a new subdomain with nothing pointing at it sits near the bottom of that queue.
Two things follow from this that surprise people:
A subdomain does not inherit the parent domain's standing. blog.example.com is treated as a separate site from example.com. Whatever reputation the main site earned does not transfer automatically.
Sitemaps and IndexNow announce existence; links vouch for value. We had done both announcements correctly, and they were not the constraint. The constraint was that nothing on the open web indicated the pages were worth spending crawl budget on.
The decisive check was embarrassingly simple: the company's main website did not contain a single link to its own blog.
What Fixed It
- A link from the main site to the blog, using descriptive anchor text rather than "click here." This is the authority signal, and it is the one that had been missing.
- "Request indexing" in Bing Webmaster Tools' URL Inspection screen. This bypasses the crawl-budget queue and forces a fetch. It is rate-limited to roughly ten URLs per day, so spend them on your most important pages.
- IndexNow, which we already had wired to fire automatically on publish. Useful, but on its own it was not sufficient — it announces, it does not vouch.
Two Related Traps Worth Checking
Soft 404s on Cloudflare Pages. If a Pages site has no root 404.html, every non-existent path returns the home page with HTTP 200 instead of a 404. Deleted or merged article URLs keep answering successfully, and the engine sees many different addresses serving identical content — a duplicate-content signal on a site that is already struggling for crawl budget. Adding a 404.html makes Cloudflare serve a real 404 for unmatched paths.
Checking your Bing index without getting blocked. Running a site: query directly on Bing tends to hit a CAPTCHA, and Bing shows fuzzy unrelated results with an "about N" count when the real answer is zero — so the count misleads you. DuckDuckGo uses the Bing index and answers site: queries without the challenge.
The Takeaway
When you see a large gap between ChatGPT and the other engines, check index status before you touch the content. In our experience the order of investigation should be:
- Is the site registered in Bing Webmaster Tools, including the subdomain separately?
- In URL Inspection, what is the per-URL status — discovered, crawled, or indexed?
- Does anything link to the pages, starting with your own main site?
- Does the site return a real 404 for paths that do not exist?
Only after those four does content quality become the useful question.
FAQ
Why does Bing matter for ChatGPT specifically?
ChatGPT's web results have historically been served through Bing's index. Gemini and Perplexity use different retrieval, which is why they were able to surface the same pages while ChatGPT could not.
How long does the fix take to show up?
We do not have a reliable number to give you, and we would rather say so than invent one. Requested indexing is fast to crawl; whether that turns into mentions depends on the content competing on its merits afterward.
Does this mean subdomains are a bad idea?
Not necessarily — they are simple to set up and easy to hand back to the client. It does mean a subdomain starts from zero and needs an explicit link from the main site on the day it launches, rather than months later.
RECO AI Lab measures how ChatGPT, Gemini and Perplexity answer local-business questions in English and Korean, publishes bilingual content on the client's own domain, and tracks index status as part of the service. Methodology and pricing are public at recoailab.com.