What the ChatGPT Citation Collapse Actually Means for Local Service Businesses

By April 2026, ChatGPT was showing source links in only about 22% of its responses. The other 78% contained no clickable source, no footnote, no visible signal of where the answer came from. Business owners reading about this correctly concluded something had changed. Most drew the wrong conclusion about what to do next.

What actually happened

ChatGPT's shift away from source footnotes was documented in a Vismore.ai study of 750 responses in April 2026. Between February and April, the rate of responses including explicit source links dropped 86-94%. The "Sources" section -- the block of clickable links ChatGPT displays after web-search responses -- effectively disappeared for most query types.

The cause was a behavioral shift in how ChatGPT handles search. ChatGPT now answers more queries from parametric knowledge (training data), reserving live web search for queries it genuinely cannot answer from memory. When training data is sufficient, there is no source to cite. No search happened.

The business owner reading about this reasonably concludes: ChatGPT isn't discovering new businesses, so AI visibility work is pointless. That conclusion is wrong in two specific ways that matter for the fix plan.

Mentions and citations are not the same metric

In our June 24, 2026 investigation into how this shift affects Sourcepull's audit methodology (`methodology-recs/2026-06-24-chatgpt-audit-mention-vs-citation-framing.md`, session 57), we documented an important distinction the press coverage mostly missed.

There are two different things being measured:

**Footnote citations:** The percentage of responses where a business appears as a linked source in the "Sources" section. This is the metric that collapsed 86-94%.

**Brand mentions:** The percentage of responses where a business name appears in the answer text -- "Smith Plumbing is well-regarded in Hamilton" -- no link required. This is what audits actually measure, including ours.

ChatGPT mentions businesses approximately 3x more often than it cites them with a source link. The footnote collapse reduced the footnote metric, not the mention metric. A business that appears in ChatGPT's answer text without a footnote has still received a recommendation. For a homeowner deciding who to call, a mentioned name without a link is usually enough.

Most of the alarm about ChatGPT citation collapse is people conflating two different measurements.

Local queries don't follow the platform average

The 78% zero-citation response rate is an average across all ChatGPT query types. It obscures a specific category where the math is different: local recommendation queries.

"Best HVAC contractor in Hamilton, Ontario" cannot be answered from training data. ChatGPT has no memory of who the best local plumber in Dallas is. For these queries, ChatGPT rewrites the question as a Bing search, retrieves the top 20-30 results, and synthesizes an answer from live web content. Web-searched responses include source footnotes.

This means the 78% zero-citation rate does not apply equally to local service recommendation queries. For "emergency electrician near me" or "best dentist in Scarborough," ChatGPT is still running web search, still showing sources, and still surfacing the businesses that appear in that Bing result pool.

No published study has measured the zero-citation rate specifically for local recommendation queries. But the mechanism is not ambiguous -- it follows directly from how the system is built. Local queries trigger web search because training data cannot answer them. Web search produces sources. As we noted in the June 24 rec, "Bing Places and Bing organic ranking affect what appears in ChatGPT's source list for local queries."

If your business is a local contractor, dentist, HVAC company, or restaurant, you drew a conclusion from platform-wide averages that does not apply to your query type.

The actual citation hierarchy for local services

If local queries trigger ChatGPT web search, the next question is: what does ChatGPT find when it searches?

In our June 23, 2026 investigation into fix plan priorities for local home services (`methodology-recs/2026-06-23-reddit-not-local-smb-fix-plan.md`, session 56), we reviewed Q4 2025 citation volume data across ChatGPT and Perplexity for local contractor queries. The hierarchy was not close:

| Source | Q4 2025 citations | |---|---| | Yelp | 512,680 | | BBB | 149,710 | | Angi | 145,633 | | Thumbtack | 56,004 | | HomeAdvisor | 33,582 | | Nextdoor | 10,308 | | Reddit | Not ranked |

Yelp is cited roughly 3.5x as often as BBB for local services. Reddit -- which has significant AI citation share in other categories -- does not appear in the top 6 for local contractor queries. The citation pool for local service recommendation queries is directories, not communities.

This matters because the directories dominating that list are all indexed in Bing. When ChatGPT performs a web search for "HVAC contractors in [city]," Yelp's result for that city is almost certainly in the candidate pool. A business with an unclaimed Yelp profile -- or one with one review and an outdated address -- is already behind the businesses that populated their profile years ago.

What the platform architecture tells you about fix priorities

The Q4 2025 local services data aligns with what we documented from the Yext 2026 AI Visibility study (17.2 million citations across ChatGPT, Perplexity, and Gemini) in our May 2026 platform citation research (`knowledge/platform-citation-behaviors.md`, session 15). Each platform has a structurally different source pool:

ChatGPT draws approximately 49% of its citations from third-party directories. Gemini draws approximately 52% of citations from brand-owned websites. Perplexity favors niche expert directories specific to the business category, not the general-purpose directories that drive ChatGPT.

For a local service business, this maps to a prioritized fix sequence.

Yelp is non-negotiable. The Q4 2025 data (512,680 citations) confirms it. A claimed Yelp profile with accurate information and 10+ reviews is the single highest-leverage action for local presence on both ChatGPT and Perplexity. The gap between Yelp and the next-ranked source is not close.

After Yelp, BBB and the category-appropriate directory (Angi for home services, Healthgrades for healthcare, Clutch for agencies) are the next actions. Consistent NAP data across these directories -- the same name, address, and phone number in every listing -- strengthens the entity signal AI systems use to match query results to business records.

Brand site work and schema markup are worth doing, but they are primarily Gemini levers. A business spending time on FAQ page schema while its Yelp profile is unclaimed has its priorities inverted.

What the collapse actually changed

The February-April source footnote collapse changed one thing: it made the footnote metric less useful as a proxy for overall business visibility on ChatGPT. It did not change the local query mechanism, did not remove Yelp from ChatGPT's citation pool, and did not reduce the importance of directory presence for local service categories.

If your business is not appearing in local ChatGPT recommendations, the cause is almost certainly not the February-April footnote shift. It is one of the same issues it was before: missing directory presence, NAP inconsistency, or insufficient Bing indexing to surface in the web search that feeds local ChatGPT responses.

A Signal Check on your domain will show you which platform is failing and whether the fix is a directory claim, a data correction, or something deeper.