75 Percent Type Keywords. Why the Long-Conversation AI Search Myth Is Costing Contractors Real Calls.

Everything you have been told about how homeowners use ChatGPT is wrong. A 177-session study just dropped and it nukes most of what trade marketers have been selling. 75 percent of users still type keywords. 45 percent of sessions are a single prompt. The average user looks at 3.7 businesses and moves on. If your contractor brand is not in the first answer, you do not get a second chance.

Read that again. The "long conversational AI search" narrative is mostly a myth. The behavior looks closer to Google in 2003 than the demos in keynote videos.

What the Sagapixel Study Actually Found

54 real people. 177 ChatGPT sessions. Tasks that ranged from finding an emergency plumber to picking a chiropractor. Search Engine Land, ALM Corp, and Coalition Technologies broke the data down. Five facts every contractor needs in their head right now.

• 75 percent typed keywords. Not full sentences. Not natural conversation. Things like "emergency plumber Brooklyn" and "chiropractor near me." The same shorthand they would type on Google.
• 45 percent of sessions were one prompt. The user asked once, got a list, picked from it. No follow-up. No refinement.
• 2.1 prompts on average per session. Even when there were follow-ups, sessions were short.
• 3.7 businesses examined per session. Only 27 percent of users looked at just one business. Three out of four explored multiple recommendations.
• 38 percent of discoveries got attributed to Google, not ChatGPT. 20.5 percent of users verified ChatGPT picks on Google before calling. 17.6 percent clicked through to a website from Google after first finding the brand on ChatGPT. The lead tracker says Google. The actual discovery was ChatGPT.

That last one is why most of you think AI search is a tiny channel. It is not tiny. It is invisible to your current attribution. The phone rings, the customer says "I found you on Google," and you keep buying Local Service Ads while ChatGPT is doing the actual work.

What This Means for Trade Contractors

Three things change about how you should run AI visibility this quarter. None of them are theoretical.

First, keyword research is not dead. The trade marketers telling you to throw out your Google keyword playbook are wrong. The same "emergency plumber [city]," "AC repair [neighborhood]," and "panel upgrade [zip]" terms that ranked you on Google are the exact phrases users type into ChatGPT for local services. Keep your keyword research. Layer GEO best practices on top of it. Do not replace one with the other.

Second, you have to be in the consideration set, not the top spot. Google rewards the number one ranking. ChatGPT rewards inclusion. If a homeowner sees five plumbers in their first answer and looks at 3.7 of them, you need to be one of the three or four that gets picked apart. Differentiation in the snippet — license number, response time, specific brand certifications, a real review quote — is what makes a user choose you over the four others on the screen.

Third, your attribution model is lying to you. If 38 percent of AI-discovered customers credit Google when they call, your "Google did it" data is overstating Google and erasing AI. Add a one-line "How did you hear about us" question to your intake form. Put a separate "ChatGPT or another AI" option. Train your CSRs to probe past the first answer. Within 60 days you will have a real read on your AI channel.

The 30-Day Visibility Build

Here is the order of operations that moves the needle. Not theory. The actual sequence.

• Week one. Run the queries. Take your top three services, your top three zip codes, and run "emergency [trade] [zip]" through ChatGPT, Gemini, and Perplexity. Screenshot the lists. Note which competitors are appearing. That is your real opponent map.
• Week two. Fix the data layer. Google Business Profile categories, attributes, hours, services, photos, Q and A. Your website schema for LocalBusiness, FAQPage, and Service. The AI does not invent details about you. It pulls from the cleanest, most consistent data it can find.
• Week three. Ship the answer pages. One page per top service. Question-based H2s. FAQ schema. Real pricing ranges. Real timeline expectations. Real local references. The page that answers the homeowner's actual question gets cited in the AI summary.
• Week four. Operationalize reviews. Texted-after-job review request. Public response within 24 hours. Sentiment loop back to operations. Reviews carry more weight in AI-mediated local search than they did in classic search because they substitute for trust signals the model cannot see directly.

The Honest Bottom Line

AI search for local services is not a brand-new game with brand-new rules. It is the old game played on a smaller stage with less margin for error. Keywords still matter. Top-three placement still matters. Reviews still matter. Schema still matters. What changed is that the user only sees a handful of businesses before they pick. The funnel got shorter and the stage got smaller.

The contractors who win 2026 are the ones who treat AI search as Google with the volume turned down and the trust signals turned up. Be in the answer. Be specific in the snippet. Track the channel honestly. The rest is noise.