Your call logs are a better content strategy than any keyword tool.
That's not an opinion—it's a structural problem with how most local service businesses approach their websites. A keyword tool tells you that people search for "AC repair Coral Springs." It won't tell you that someone opened ChatGPT and typed something like: imagine a customer asking, "The same company serviced my AC two months ago and now it's not cooling again—do I call them back or is this something different?" That second version has a timeline, a constraint, a decision to make, and genuine buying intent. Keyword tools strip all of that context away.
AI assistants—the chatbots that answer questions conversationally, like ChatGPT or Gemini—don't strip anything away. They answer the whole situation. So if your website speaks to whole situations rather than just short search phrases, you have a real shot at being the source those assistants pull from when your potential customers ask.
The question is: where do you find those full-situation questions?
Start With the Words Coming Out of Your Customers' Mouths
Sales calls, pre-booking inquiries, support tickets, live chat logs—these are your single best source, and most business owners have never looked at them as website content material.
When someone calls before they book, they describe their situation out loud the same way they'd describe it to an AI assistant. The phrasing, the constraints ("I had someone else look at it already"), the timeline ("it just started yesterday"), the hesitation ("I'm not sure if this is even your kind of job")—all of it is there, unfiltered.
Most call and scheduling tools let you export transcripts—a written record of what was said on a call. Pull the last three months. Then paste those transcripts into an AI assistant (with any customer names and identifying details removed first—more on that in a moment) and use a prompt like this:
From these transcripts, list every question the customer asked or implied. Rewrite each one the way they'd phrase it to an AI assistant, keeping their specific details—what they've already tried, their timeline, their situation. Group similar questions and note how many calls each type came up in.
That last instruction matters more than it looks. If the same type of question comes up in five separate calls, that's the closest thing to proof of demand you'll find for a phrase that will never appear in a keyword tool. It's far more convincing than a gut feeling when you're deciding what to put on your website.
Before you do this: strip out customer names, addresses, and any identifying details before pasting anything into a third-party tool. Also review your call recording consent policy to make sure this kind of internal use is covered.
One more thing: calls tend to skew toward people who are already close to booking. Add your support tickets and pre-sale inquiry emails into the same exercise to catch the earlier-stage questions too—the people who are still figuring out whether they even need you.
Fill the Gaps With Forums and Reviews
Your own calls only capture people who already found you. Forums fill in everyone else.
Reddit is the obvious starting point, but Facebook groups, neighborhood apps like Nextdoor, and review sites work just as well depending on your trade. The value is usually in the opening post, where someone explains their full situation before they even know the right question to ask. The replies matter too—they show which follow-up questions a stranger, or an AI, would need answered before they could actually help.
Search the platform for your trade combined with a situation your customers commonly describe. If you do mold remediation, try searching Reddit for "mold found after water damage" and read how homeowners explain their situation before they've talked to anyone. Copy the most relevant threads into a document and run the same extraction prompt you used on your call transcripts.
The goal is one unified list of real questions—from calls, tickets, and forums—all in the same format, all written the way a real person in a stressful situation would actually say them.
See How AI Actually Breaks a Question Down
Here's something worth understanding about how AI assistants work: when you ask one a detailed question, it doesn't just find one page and quote it. It tends to break the question into several smaller sub-topics—specific angles of the question—and look for sources that address each one separately. Researchers who study AI search behavior have documented this pattern across the major AI tools.
You can see this yourself without any technical knowledge at all.
Take one of the real questions from your call logs and paste it into ChatGPT or Gemini. Watch which angles the response covers. Does it address cost? Timeline? What to try first versus when to call a pro? Which warning signs to watch for?
Now open your website and ask honestly: does any page actually address this question at that level of specificity?
If the answer is no, that's a gap. If it's "kind of, but only in one sentence," that page needs more depth on the specific scenario your customers are actually in.
Think of the AI's response as a coverage checklist—a list of the angles your page would need to address to be genuinely useful as a source. It's a planning tool, not a guarantee of anything, and the response can vary each time you run it. But it's a fast way to see what's missing from your site compared to what a customer actually needs to know.
Check What's Already Reaching You
Google Search Console is a free tool from Google that shows which search terms are already sending people to your website. It's useful here as a reality check, not as the starting point.
Longer, conversational queries—the kind that sound like a real person describing a situation—do occasionally appear in the Performance report inside Search Console. Filter for queries that start with question words: how, what, why, should, is, can, does, when. What comes back is a small slice of what's actually happening, because Google doesn't show you queries that are too rare or too unique. Conversational questions tend to be phrased differently by every person who asks them, so the data will always undercount what's really out there.
If a question from your call logs also shows up here in some form, that's confirmation that real people are searching it and prioritize it. If a question doesn't show up, that doesn't mean nobody's asking it—it may just mean the volume is too small for Google to surface.
What to Do With All of It
At this point you'll have a messy pile of questions from several sources. The organizing step is simple: one spreadsheet, one row per question. For each row, note where it came from (call, ticket, forum), how often it came up, and what angles an AI covered when you tested it.
Prioritize by frequency. A question that came up in multiple calls, also appears in forum threads, and has some Search Console signal is the one to write about first—and to write about thoroughly, in plain language, the way your customer actually described it.
This is the content that tends to get pulled into AI answers: not the most pages, not the highest word count, but the page that actually speaks to the whole situation a real customer is in.
What to Do This Week
- Pull your last 90 days of call recordings or support tickets. Most call and scheduling tools let you export transcripts. If yours doesn't, have someone listen to five to ten calls and write down every question the customer asked before booking.
- Run the extraction prompt. Remove any customer names and identifying details, then paste the transcripts into an AI assistant and ask it to list every customer question, rewritten the way someone would phrase it to an AI, with their specific constraints included.
- Search one forum for your trade. Pick Reddit or a relevant Facebook group and search for a common problem your customers describe. Copy the opening posts from three to five threads into your working document alongside your call questions.
- Test your top three questions in ChatGPT or Gemini. Paste each one in and note what angles the response covers. Then check whether any page on your site addresses those angles at that level of detail.
- Flag the gaps. Any question that came up repeatedly in calls, also appears in forums, and isn't answered on your site in plain conversational language is your next piece of content—or a reason to revisit a page that's already there but needs more depth.