Open ten service pages from ten competitors and you will probably read the same page ten times. Same advice, same phrasing, same safe claims. This happens in many industries, especially with cookie cutter retail automotive SEO.
TL;DR: Claudia Tomina at Search Engine Land shows how the follow-up questions in Google AI Mode can reveal what your service pages are missing. The fix is specific, first-party facts only your business can supply, not more generic content.
🔁 Most Service Pages Say the Same Thing
Tomina calls it commodity content. It's a page that answers the same basic questions, in nearly the same words, as everyone else in the market.
That's what you would call commodity content.
Her example was a car accident page for a law firm in Michigan. It told readers to call an attorney quickly, explained contingency fees, and summarized Michigan's comparative-fault rule. More than a dozen competing firms said the same things, almost word for word.
That's a problem for AI search because the model has nothing distinctive to pull from your page. Your page gets summarized into the same pile as everyone else's. Then the model fills in the details buyers actually need from somewhere else.
Tomina also points to Yext's analysis of 17.2 million AI citations. It found that the models don't all choose sources the same way.
Gemini favors official websites. Claude cites reviews and social content two to four times as often as other models. Perplexity cites business-owned sites roughly 37% to 50% of the time, depending on the sector.
The takeaway is that generic content has the same weakness everywhere. Whichever source a model prefers, it has nothing special to take from you.
💬 Follow-Up Questions Show What Buyers Still Need
Tomina's key move was small. Instead of asking AI to write more content, she searched a real buyer question, like whether you need a lawyer after a car accident in Michigan. Then she read the follow-up questions Google AI Mode suggested.
Those follow-ups skipped the generic advice. They went straight to Michigan's pain-and-suffering threshold, how comparative fault changes a settlement, and what happens if the other driver is uninsured. They also asked where a local claim gets filed and when vehicle damage belongs in small claims court.
Each one is a decision the buyer has to make. Together they work as a content map for the page. She prefers AI Mode for local research because Google understands places, businesses, and the local context around a search.
🧾 The Workflow Doesn't Let AI Guess
Tomina published a 10-phase workflow on GitHub for finding these gaps. It covers auditing the existing page, writing natural-language buyer questions, mapping follow-ups, scoring gaps, and drafting content briefs.
The part worth borrowing is the labeling. Every finding is tagged OBSERVED, VERIFIED, CLIENT-SUPPLIED, INFERRED, or UNKNOWN. That keeps a plausible guess from getting published as a fact.
On the car accident page, the workflow found specifics no competitor had. It surfaced the actual local courts a case would run through, a separate small-claims threshold for vehicle damage, and the exact legal test for recovering pain-and-suffering damages.
It also refused to fill in what it couldn't know. Settlement ranges, years of experience, and case volume all came back as UNKNOWN, and each one became a question for the attorney.
🛠️ Three Ways to Fill Your Own Gaps
Search a Real Buyer Question in AI Mode
Pick the question a buyer would actually type for your top service page. Write down every follow-up AI Mode suggests. Mark the ones your page doesn't answer with anything specific.
Label Every Claim by How You Know It
You don't need the full workflow to borrow the habit. Next to each fact on the page, note whether you saw it, verified it, got it from the client, or are guessing. Anything in the guessing pile needs a source or needs to go.
Mine the Interviews You Already Have
Tomina noticed the attorney had already answered many of these questions on YouTube podcasts. She fed 13 hours of transcripts into Claude, and it closed almost all of the UNKNOWN items.
That produced the firm's fee structure, how long cases take, real verdict and settlement figures, and an anonymized case where an injury narrowly missed a legal threshold. About three-quarters of cases resolve without a lawsuit in four to six months, which is the kind of detail a competitor can't copy. A subject matter expert interview covers whatever the transcripts can't.
The finished page named specific courts and explained why they mattered. It walked through comparative-fault math with a worked example and showed the order fees come out of a payout. Because the transcripts had timestamps, each claim could link to a 30-second clip readers can check.
Hit reply
Think of your most important service page. What's one fact on it that only your business could know, and is it actually written there?
💭 One Thing I'm Thinking About
The hard part of this workflow isn't the prompts. It's admitting how much of a typical page could have been written by anyone.
In a lot of cases, the people running the business already know the answers to these buyer questions. They just never made it onto the page.
Search keeps moving toward answers, and AI tools will build those answers from whatever they can find. If the specific details live in your head and not on your page, someone else's page will supply them.
Did You Know? Because the podcast transcripts carried exact timestamps, Tomina's finished page could link every claim to a 30-second video clip, so readers and AI models could check it fast.
Till next time,
