Diagnosing ChatGPT Invisibility: A Bing-First Troubleshooting Workflow

The pattern repeats across client engagements often enough to be predictable. A brand notices that ChatGPT is not citing them for queries where they should be cited. The investigation starts with OpenAI-specific hypotheses: maybe OAI-SearchBot is blocked, maybe the robots.txt is wrong, maybe schema markup is missing. Teams spend hours diagnosing OpenAI-side issues. They find nothing meaningfully wrong. The actual cause turns out to be Bing-side: the brand fell out of Bing's index six months ago after a site migration, and ChatGPT's retrieval pipeline cannot cite content that is not in the upstream index it depends on.

The wrong default investigation order costs time and produces wrong conclusions. The right order starts with the layer that fails most often and works outward to the more specialized causes. For ChatGPT visibility specifically, that means Bing first, OpenAI second, content quality third. This playbook walks the diagnostic in the order that catches problems fastest, with the specific commands and queries that make each stage concrete.

Why The Default Investigation Order Is Wrong

When a brand asks why ChatGPT is not citing them, the natural impulse is to investigate ChatGPT-specific causes. The brand sees a problem in ChatGPT, so the investigation should focus on ChatGPT. The impulse is intuitive and almost always misguided, because ChatGPT search is not a self-contained system. It is a retrieval surface running on top of multiple upstream indexes, and a problem anywhere in the pipeline manifests at the citation level even when the proximate cause sits several layers upstream.

The architecture matters for diagnosis. ChatGPT search consults at least two indexes when it answers a query: OpenAI's own OAI-SearchBot index and Microsoft's Bing index, with the Bing layer doing the heavy lifting for most queries. Independent analysis has consistently shown citation overlap with Bing's top results around 87% across diverse query sets. If your site is absent from Bing's top results, you are absent from the foundation that ChatGPT draws on most heavily, which means the OpenAI-specific interventions you might consider (schema, robots.txt, OAI-SearchBot allow rules) cannot recover citations that the upstream layer cannot surface in the first place.

Starting the investigation with Bing instead of OpenAI usually produces a quick diagnosis. Sites well-indexed in Bing rarely have catastrophic ChatGPT visibility problems. Sites poorly-indexed in Bing rarely have meaningful ChatGPT visibility regardless of how well-tuned their OpenAI configuration is. The companion piece on why ChatGPT search uses Bing walks the architectural relationship that justifies the inverted investigation order.

The Cost Of The Wrong Order

A team that investigates OpenAI-specific causes first typically spends two to four hours doing it before realizing the proximate cause sits upstream. The same investigation, ordered correctly, takes 15 minutes to either confirm or rule out the Bing layer. The time savings compound across multiple troubleshooting sessions, which is why every diagnostic engagement we run now opens with Bing rather than OpenAI.

Stage 1: Bing Index Health

The first question is whether your pages are in Bing's index at all. The cheapest verification takes under a minute using Bing's own search engine.

Open bing.com in a browser and run a site: search restricted to your domain:

site:example.com

The result page reports an approximate count of indexed URLs and lists samples. If the count is zero or unexpectedly low (a few results when you expect hundreds or thousands), Bing has either lost most of your pages from its index or never indexed them in the first place. The investigation moves to why.

For more precise per-URL checks, Bing Webmaster Tools provides URL Inspection that reports the index status of a specific URL plus diagnostic information about why it might not be indexed. The tool requires verified property access; if you have not set that up, the companion piece on Bing Webmaster Tools setup is a 30-minute prerequisite. The URL Inspection workflow:

1. Enter a high-priority URL into the URL Inspection field.
2. Review the response. The tool reports whether Bing has the URL in its index, when it was last crawled, what the canonical URL is from Bing's perspective, and any issues it found during the most recent crawl.
3. If the URL is not indexed, check the reason: blocked by robots.txt, noindex meta tag, server error during last crawl, redirect chain, content quality issue, or "not crawled" if Bingbot has never visited.
4. Resolve the underlying issue and request a re-crawl through the same tool.

The patterns that produce sitewide deindexation are typically a few weeks to a few months old and trace back to a specific deployment, infrastructure change, or policy update. The pattern that produces partial deindexation usually traces back to content-quality signals on a specific section of the site.

The Diagnostic Reports To Run First

Inside Bing Webmaster Tools, run these reports in order:

1. URL Inspection on three top-priority URLs (homepage, top blog post, top product page).
2. Sitemap report (Submissions section). Confirm your sitemap is being processed and the URL discovery rate is healthy.
3. Crawl Information report. Look for spikes in server errors or robots.txt blocks that align with the timeline of the visibility drop.
4. Backlinks report. Healthy backlinks are an authority signal Bing weighs heavily; a sudden drop here often correlates with index inclusion issues.

The four reports together give you a complete picture of Bing-side health in roughly 15 minutes.

Stage 2: Bing Rank For Target Queries

Being in Bing's index is necessary but not sufficient. You also need to rank well in Bing for the queries you want ChatGPT to cite you on. ChatGPT's retrieval system heavily weights the Bing top results, so a page that exists in Bing's index but ranks at position 50 will rarely get cited by ChatGPT, even if the page is otherwise excellent.

The fastest way to check Bing rank for your target queries is to run them on bing.com directly. Open an incognito or private browsing window (to avoid personalization bias), enter the query, and scroll through the results. Note where your domain appears. Repeat for 10-15 of your highest-value queries to get a baseline. The patterns:

If you rank in positions 1-5 in Bing, you are well-positioned for ChatGPT citation on the query. Lack of citation is more likely an OpenAI-side or content-side issue.

If you rank in positions 6-15, you are in the "marginal" zone. ChatGPT cites these pages sometimes but not consistently. Improving rank closer to the top usually helps.

If you rank below position 15, you are functionally invisible to ChatGPT's retrieval. The fix is to improve Bing rank rather than to focus on OpenAI-specific work. Bing-side ranking factors overlap significantly with Google but emphasize some signals differently: more weight on exact-match keywords in content, more weight on backlink quantity (not just quality), more weight on social signals, slightly less weight on freshness for evergreen topics.

If you do not appear in Bing's results at all for a query you should rank for, return to Stage 1 to verify your relevant pages are actually in the Bing index. Failing to appear in results when you should be indexed indicates the page was indexed but received a penalty or rank suppression that needs separate investigation.

A Comparison Worth Running

For each high-priority query, compare your Bing rank to your Google rank. If you rank well in Google but poorly in Bing for the same query, the gap is a Bing-specific optimization opportunity that directly affects ChatGPT visibility. Bing's algorithm differs from Google's in ways that surface this kind of asymmetry, and closing the Bing gap usually yields disproportionate citation improvements relative to additional Google work on already-strong rankings.

Stage 3: OAI-SearchBot Accessibility

If Bing health and Bing rank are both healthy, the next layer to investigate is OpenAI's own OAI-SearchBot index. The bot can fail to ingest your pages even when Bing has them, which produces a citation gap on queries where ChatGPT prefers OAI-SearchBot's index over the Bing layer (typically newer, fresher, or more specialized queries).

The diagnostic order at this layer:

1. Check your access logs for recent OAI-SearchBot activity. A grep for the user agent reveals whether the bot has been visiting your site at all. If volume is near zero, the bot is being blocked or has lost interest in your site; either way, the investigation focuses on accessibility.
2. Verify your robots.txt is not blocking OAI-SearchBot. Curl your robots.txt and look for User-agent: OAI-SearchBot disallow rules or User-agent: * disallow rules that would catch it. The companion piece on reading OAI-SearchBot crawl logs walks the verification in detail.
3. Verify your CDN bot management is not blocking OAI-SearchBot. CDN dashboards (Cloudflare, AWS WAF, Fastly, Akamai) all have bot management features that can challenge or block AI crawlers indiscriminately. Check the dashboard settings, particularly any features labeled "AI bot management" or "super bot fight mode."
4. Confirm the bot can reach your origin successfully. If OAI-SearchBot fetches show 4xx or 5xx responses in your access logs, the bot is being rejected at some layer between the public URL and your origin server.
5. Confirm the OAI-SearchBot fetches return useful content. A fetch that succeeds but returns an empty page (client-side rendering issue), a paywall (auth issue), or generic boilerplate (caching issue) does not produce indexable content.

If all five points check out, OAI-SearchBot is functioning as expected and the citation gap, if any, sits at the content layer rather than the accessibility layer.

The Quick OAI-SearchBot Confirmation

A reproducible single-line confirmation:

grep -i "OAI-SearchBot" /var/log/nginx/access.log | tail -20

If you see recent fetches with 200 responses to article-level URLs, the bot is reaching your site successfully. The investigation moves up the stack.

Stage 4: Content And Structural Issues

If both upstream indexes can reach and ingest your pages but ChatGPT still does not cite you, the remaining causes live in content and page structure. Several patterns recur often enough to be worth checking systematically.

JavaScript-dependent content is the first pattern. OAI-SearchBot does not execute JavaScript by default. Pages whose content only assembles after client-side rendering are invisible to the bot even when the URL fetches successfully. The diagnostic is to view the source of the page or fetch it with curl and inspect what the bot would see. If the article body, product description, or pricing information is missing from the raw HTML and only appears after rendering, server-side rendering is the fix.

Schema markup absence is the second pattern. Pages with JSON-LD describing the content type (Article, Product, FAQPage, BreadcrumbList) are easier for AI engines to extract and classify. Pages without schema can still be cited, but the citation rate is lower. The diagnostic is to check the page source for application/ld+json script blocks and verify the schema validates against the relevant Schema.org type.

Heading hierarchy issues are the third pattern. AI engines use heading structure to identify which sections of a page answer which sub-questions. Pages with no H2 elements, jumping from H1 to H4, or wrapping all content in generic divs without semantic headings are harder for the engine to chunk meaningfully. The fix is to restructure with clean h2/h3 hierarchy.

Authority and external signal patterns are the fourth. AI engines look for external validation (citations from authoritative sources, mentions in domain-relevant publications, presence in Wikipedia or Wikidata) when scoring trustworthiness. A page whose content is excellent but whose domain has weak external signals can struggle to earn citations against competitors with stronger external footprints.

Content depth and originality are the fifth. AI engines favor pages that include original data, specific examples, named entities, and concrete details over pages that summarize generic information available across many sources. The content patterns that earn ChatGPT citations are documented in our companion playbook and worth reviewing in detail when content-quality issues are suspected.

The Per-Page Diagnostic

For specific high-priority pages that should be cited but are not, run a five-question audit:

1. Does the raw HTML (without JavaScript) contain the article body?
2. Is there valid JSON-LD schema describing the page content type?
3. Are headings h1 -> h2 -> h3 in sequential order without skipping levels?
4. Does the page include statistics, original data, named entities, or specific examples that distinguish it from generic content?
5. Does the page have external citations or mentions from authoritative sources in the topic area?

If any of the five answers is no, that is a candidate cause for the citation gap and a candidate intervention.

Stage 5: The Citation Test Itself

After working through the upstream stages, the final verification is to run citation tests directly in ChatGPT. The test confirms whether the work at the previous stages has moved the citation outcome, which is the only metric that ultimately matters.

The protocol:

1. Identify 10-15 queries that buyers in your category typically ask. Mix transactional, comparative, and research queries.
2. Run each query in ChatGPT (not ChatGPT search, just the main chat surface with web access enabled if possible).
3. Note whether your domain appears in the source citation list.
4. Note whether competitor domains appear and which ones.
5. For queries where competitors are cited and you are not, click through to the cited competitor pages and examine what they do differently from your equivalent pages.

The output is a citation matrix: queries by domains with cited/not-cited marked. Run the test monthly to track changes over time.

For more rigorous measurement, use the Perplexity API, the OpenAI API with web search enabled, or the Gemini API with grounding enabled to programmatically test citation rates across larger query sets. Helper scripts that automate the cross-engine test are documented elsewhere in the playbook library. The principle is the same regardless of implementation: empirical citation testing is the ground truth for whether your AI visibility work is paying off.

When To Re-Run The Diagnostic

After making changes based on the diagnostic findings, allow 2-4 weeks for the changes to propagate through Bing's index and OAI-SearchBot's index. Re-run the citation tests at that point to measure impact. The propagation window for individual page changes is shorter than the window for site-level structural changes; expect quick feedback on per-page interventions and longer windows for sitewide overhauls.

Common Anti-Patterns This Diagnostic Catches

Across the dozens of "why is ChatGPT not citing us" engagements we have worked through, six anti-patterns recur often enough to be worth flagging explicitly:

1. The brand blocked GPTBot in robots.txt and assumed that explained the visibility loss. GPTBot is the training crawler; blocking it does not affect ChatGPT search citations. The actual cause is usually upstream of OpenAI entirely.
2. The brand assumed Bing was unimportant and never set up Bing Webmaster Tools. The Bing layer drives most ChatGPT citations, and being absent from the Bing index produces the visibility gap regardless of how well-tuned the OpenAI configuration is.
3. The brand ran a site migration and lost Bing indexation in the process. The pages still exist on Google, but Bingbot has not recrawled the new URL structure. Submitting an updated sitemap to Bing and confirming index inclusion fixes most of these.
4. The brand has a CDN bot management feature configured to block AI crawlers indiscriminately. OpenAI's published IP ranges and user agents get caught in the same rules that block scraper traffic. The fix is at the CDN dashboard, not the publisher's site.
5. The brand publishes content that only assembles after client-side JavaScript. OAI-SearchBot sees an empty page, Bingbot may also struggle, and the content is invisible to AI engines even though it renders correctly for users. Server-side rendering or static generation closes the gap.
6. The brand publishes content that lacks specific facts, original data, or named entities. AI engines preferentially cite pages with extractable, attributable claims over pages that summarize generic information. Strengthening the content with statistics, examples, and quoted sources improves citation rates measurably.

Each anti-pattern is fixable. The right intervention depends on which stage of the diagnostic the issue surfaces in, which is why the ordered workflow above is more useful than ad-hoc investigation.

The Time Budget For Each Stage

Stage 1 (Bing index health) takes 15-30 minutes for an initial check. Stage 2 (Bing rank) takes 30-45 minutes for 10-15 queries. Stage 3 (OAI-SearchBot accessibility) takes 30-60 minutes including CDN dashboard review. Stage 4 (content and structural) takes 60-90 minutes per high-priority page. Stage 5 (citation testing) takes 30-45 minutes for the test set, with additional time for analysis. The full diagnostic runs in a single workday for most brands.

Frequently Asked Questions

Should I run this diagnostic even if ChatGPT is citing me sometimes?

Yes. The diagnostic is useful for both fully-invisible and partially-visible sites. A brand cited for some queries but not others has an opportunity to improve the citation rate on the missing queries, and the diagnostic surfaces the gaps. Running the workflow quarterly catches incremental issues before they accumulate into larger visibility losses.

What if I rank well in Bing but ChatGPT still does not cite me?

The remaining cause is usually one of: OAI-SearchBot accessibility (Stage 3), content quality (Stage 4), or topic-specific issues where ChatGPT prefers different sources than Bing for the particular query category. Wikipedia, aggregator sites, and high-authority publishers sometimes appear as ChatGPT citations even when they do not top Bing's rankings, which means competing with them requires authority signals that go beyond keyword rank.

How long does the full diagnostic take?

For an initial run, 4-8 hours of work depending on site complexity. For a brand new to the workflow, expect to spend more time orienting in Bing Webmaster Tools and reading server logs. For an experienced team, the diagnostic compresses to 2-3 hours and can be run quarterly as a maintenance routine.

Can I delegate this to my agency or run it in-house?

Either works. The skills required are: comfort with server log analysis, familiarity with both Google Search Console and Bing Webmaster Tools, basic understanding of HTTP and JSON-LD, and patience for empirical testing. Most SEO agencies and reasonably-staffed in-house teams have the relevant skill set. The work is more process than craft, which makes it well-suited for delegation once the workflow is documented.

What if the diagnostic shows everything is healthy but I am still not cited?

A genuine "all healthy but not cited" outcome is rare but possible. The remaining causes are usually subtle: domain authority below the threshold ChatGPT considers for the query category, content that is well-structured but lacks distinctive value, or temporary issues with OpenAI's retrieval system that resolve on their own. The right response is to continue strengthening the content and authority signals while monitoring citation rates monthly, and to re-run the diagnostic if the gap persists more than 90 days.

The right diagnostic order for ChatGPT invisibility is upstream-first. Bing before OpenAI. Index before rank. Accessibility before content. The order works because the failure modes cluster in predictable ways, and the time investment in the wrong order costs hours that the right order saves. Walking the stages in sequence produces a diagnosis within a workday for most sites, and the resulting interventions are typically tractable engineering or content work rather than mysteries that resist resolution.

If your team wants the full diagnostic run as a structured engagement (with the Bing audit, the OpenAI accessibility check, the content review, and the citation testing all instrumented), that work sits inside our generative engine optimization program. The workflow is well-understood. The interventions are mostly mechanical. The brands that close the gap and stay closed are the ones who instrument the diagnostic and re-run it on a quarterly cadence.