The cross-engine overlap rate is flat by category. One engine pair is not.

An earlier post found that only 24% of the domains cited across our four AI engines show up on more than one engine's reading list, and that the number barely moves panel to panel. It did not check whether the category itself changes that picture. We reran the same calculation split by project management, HR, and customer support software. The headline rate holds flat everywhere. One specific engine pairing does not.

The overlap rate is close to identical in all three categories

Same method as before: for each of the 270 live audit panels, we pulled every citation URL from every real, non-error response, normalized it to a hostname, and excluded Gemini's opaque vertexaisearch.cloud.google.com redirects. That gives a set of distinct domains per engine per panel, and for each domain we counted how many of the four engines cited it.

20,313 plus 6,345 plus 4,254 domain-panel pairs adds up to the same 30,912 the blended post reported, and the 4,817 plus 1,552 plus 1,057 overlap counts add up to the same 7,426, confirming this is a category split of the identical dataset, not a new sample. The per-panel median overlap share also barely moves: 23.6% in project management, 25.2% in customer support, 25.7% in HR software, each with a similar P25-P75 spread of roughly 21-28 points. Whatever category a panel covers, about one domain in four ends up on more than one engine's list. Category is not the variable driving that number.

But the ChatGPT-Claude pairing reorders itself by category

The blended post reported six pairwise Jaccard overlaps and found every pair involving ChatGPT sitting well below the other four, ChatGPT's weakest link with anyone. Splitting those same six pairs by category shows that framing only holds for one category.

In project management software, ChatGPT vs Claude is the single weakest pairing of the six, at 4.2%, behind both of ChatGPT's other pairings. In HR and customer support software, the same pairing jumps to 11.3% and 12.0%, ahead of ChatGPT's other two pairings and within range of Perplexity vs Claude. The three ChatGPT-Gemini and ChatGPT-Perplexity pairings stay roughly where the blended post put them across all three categories. It is specifically ChatGPT and Claude's shared reading list that looks different depending on which category you check: barely overlapping in PM software, meaningfully overlapping in HR and CS.

Summed back up, the three categories' shared-domain counts for this one pairing (163 in PM, 99 in HR, 439 in CS) add to 701, matching the blended post's ChatGPT-Claude figure exactly, and the category union counts (3,892, 880, 3,667) add to 8,439, also an exact match. The reordering is inside the same verified numbers, not a different cut of the data.

Part of the story is ChatGPT's coverage, but not all of it

ChatGPT's real-world API reliability is not flat across categories, documented in an earlier post: its live-call error rate ran 83.2% in project management and 88.0% in HR software but just 0.3% in customer support. That shows up here as panel coverage: ChatGPT had usable citation data in only 57 of 173 project management panels (33%) and 15 of 39 HR panels (38%), but in all 58 of 58 customer support panels (100%).

Customer support's near-full ChatGPT coverage plausibly explains part of why its ChatGPT-Claude overlap is higher there than in project management. It does not explain HR software's jump. HR's ChatGPT panel coverage (38%) is close to project management's (33%), yet HR's ChatGPT-Claude overlap (11.3%) is nearly three times project management's (4.2%). Thinner data availability is a real caveat on the customer support number specifically. It is not the whole explanation for the category difference.

Why this matters for a visibility strategy

If you are prioritizing which engine-specific source gaps to close first, the blended "ChatGPT shares the least with everyone" rule is a reasonable starting assumption in project management software. In HR and customer support software, it understates how much a ChatGPT-facing source win might also help on Claude, and vice versa. The category your brand competes in changes not just your visibility floor, documented in an earlier post, but which cross-engine shortcuts are actually available to you.

Methodology

Data drawn from the same 270 live, search-grounded audit panels used in the original cross-engine overlap post (173 project management, 58 customer support, 39 HR software brands), each run across four AI engines: ChatGPT with web search, Gemini with grounding, Perplexity Sonar, and Claude with web search. For every real (non-error) response, we pulled the citation URL list, normalized each to its hostname, and excluded Gemini's vertexaisearch.cloud.google.com redirect wrapper. For each panel we built the set of distinct cited domains per engine, then split panels by their config.category field before recomputing both the per-domain overlap share and the pairwise Jaccard similarity between each pair of engines' domain sets, pooled within each category. Panel coverage by engine and category: ChatGPT 57/173 (PM), 15/39 (HR), 58/58 (CS); Gemini, Perplexity, and Claude each covered 96-100% of panels in every category. All shared-domain and union counts were cross-checked against the original blended post's published totals and matched exactly, confirming the same underlying dataset re-split, not a new sample. Numbers recomputed fresh from results.jsonl this session. No development-rail or fixture data is included; all responses came from live engine calls. Data collected June-July 2026.