Which Trades Do AI Engines Recommend Agencies For?

Which source do AI engines cite most for each trade?

The table lists the top two cited domains per trade, with the number of times each was cited across all ChatGPT and Perplexity runs in the window, and baadigi.com's own count for the same trade. Counts are URL citations, so a domain that an engine cites on every run of the same prompt is counted every time.

Source: BaaDigi AI citation tracker, ChatGPT and Perplexity, September 1 to October 27, 2026. Twelve of the 27 tracked trade categories are shown; the remaining 15 (27 − 12 = 15: appliance repair, auto detailing, concrete, construction, fencing, home builder, locksmith, paving, pool, pressure washing, restaurant, septic, solar, tree service, window and door) are in the tracker but not in this table. BaaDigi is excluded from the third-party ranking and shown separately in the last column.

Which trades did Reddit and Thumbtack lead?

Reddit led roofing, HVAC, plumbing and painting in our prompt sample. Electrical, landscaping and remodeling went to thumbtack.com, with Reddit second in each.

Our first study in August, We Ran 1,429 AI Searches. The Engines Disagree, also had reddit.com as the most-cited domain overall, with 657 citations. That earlier study covered three engines (ChatGPT, Perplexity and Google AI Overviews); this one covers two, so the totals are not directly comparable.

In other categories the top domain was neither Reddit nor Thumbtack. bluecorona.com led insulation with 80 citations and treetraction.com led tree service with 221. We report the observed counts and do not infer why they came out this way.

How often does BaaDigi get cited, trade by trade?

We publish our own number because a study that hides the publisher's row is not worth citing. Across all 13,113 ChatGPT and Perplexity runs, the rate at which an answer cited baadigi.com varied from 57.0% on window and door prompts to 0.0% on insulation. The second table shows a selection.

Read this table as a measurement of our own pages and prompts, not of the trades. Our trade roundups live at our agency comparison pages. Roofing, HVAC and plumbing also include several "is Angi worth it for X" prompts that rarely cite anyone's agency page, which pulls those three rates down. It is a prompt-mix effect, not evidence that AI dislikes roofing content. Page presence alone does not explain the spread: our restoration comparison page was first cited on October 2, 2026, and restoration still came in at 5.0%.

Which kinds of questions get an agency cited at all?

Grouping the 310 prompts by what they ask shows a wide spread. The trade agency roundup prompts ("best marketing agencies for plumbers") cited baadigi.com in 52.0% of 1,358 runs. Neutral comparison prompts ("agency A vs agency B") hit 47.9% of 384 runs. At the other end, definitional prompts ("what is a map pack") cited us in 3.8% of 658 runs, and new-contractor buyer-stage prompts in 7.4% of 474.

The lesson is not about us. It is that AI engines cite pages whose shape matches the question. A roundup question pulls a roundup page. A definition pulls a glossary. The vertical cost and spend prompts cited us in 29.3% of 1,328 runs. If you want an engine to name your business for "best roofer in Fort Worth," you need a page that reads like an answer to that exact question, with a figure in it, on your own domain. You can check what the engines currently say about your business with the free AI search visibility checker.

How did we measure this?

Source: BaaDigi's AI citation tracker, a script that sends a fixed list of prompts to ChatGPT and Perplexity through their APIs and stores every URL each engine cites with its answer. Window: September 1 to October 27, 2026; 24 run days, 310 prompts, 13,113 engine-runs with no error. Trade is assigned by the first trade word found in the prompt text; 154 prompts (6,580 of the 13,113 runs) contain no trade word and are excluded from the trade tables. The prompts are not all the same age: some buckets were seeded on September 30 and October 1 and have about 11 run days, against about 24 for the older ones, so rates are not age-matched. Domain counts strip protocol, www, paths and tracking parameters, so a page cited with and without a UTM tag counts as one domain.

What this can support: statements about which domains these two engines cite, in what proportions, for this set of 310 BaaDigi-written prompts. What it cannot support: anything about search volume, how many real people ask these questions, clicks, leads or revenue. There is no demand signal in the tracker. It is a convenience sample, not a random one: the prompts were authored by BaaDigi to track BaaDigi, so our own rank is a product of prompt selection and dozens of prompts that target pages we built. It is not an industry benchmark. Rates are per engine-run, not per person; the same prompt is re-asked on every run day, so a page one engine likes gets counted up to roughly 24 times. Most trades rest on 4-prompt samples, so a single prompt answered with a listicle can put a niche domain in first place. Engine behaviour changes without notice. Treat every number as a snapshot of the September–October observation window.

What should you do with this?

• If you are a roofer, HVAC, plumbing or painting contractor: Reddit was the most-cited domain for your trade in our sample. Publish pages that answer the exact question with real prices and real outcomes, and measure per engine, because a result in Perplexity says nothing about ChatGPT.
• If you are in septic, restoration, insulation or tree service: the top domain in our sample was neither Reddit nor Thumbtack. Check which domains your engines cite for your trade before you build a page.
• If you sell marketing to contractors: the 52.0% citation rate on roundup prompts versus 3.8% on definitions says to publish comparisons and cost tables, not glossaries.
• If you want a single blended "AI visibility score": do not buy one yet. The per-trade and per-question spread here is too wide for one number to mean anything.

We plan to re-run this index and publish the deltas. The next cut adds Google AI Overviews once its citation extraction is stable enough to count.