Stop Generating Blog Posts From Keyword Lists. Your Search Terms Report Is a Better Editorial Calendar.
Michael Sacca•Every AI blogging guide starts in the same place. Open a keyword research tool. Type a seed phrase. Export four thousand rows. Sort by volume, filter by difficulty, pick twenty winners. Hand them to a writer, human or model, and wait for the traffic.
It feels like strategy. It is actually shopping.
The tool is estimating what the entire world searches for. The entire world includes every competitor you have, and they typed the same seed phrase into the same tool and exported the same four thousand rows. You are not finding topics. You are splitting them.
Meanwhile, in a tab you already pay to keep open, Google is keeping a list of the exact questions your buyers typed into a search box moments before some of them gave you money.
The keyword list is a consensus document
Ahrefs and Semrush both model the same public web. They crawl the same results pages, estimate the same volumes, score the same difficulties. The entry tier of a serious suite runs north of a hundred dollars a month, and what it sells you is a consensus: the average of what everyone, everywhere, might want.
Same inputs in, same posts out. When your outline and your competitor's outline both descend from the same keyword export, the articles converge, and interchangeable content is exactly what rankings and AI answers have both learned to skip. We made the long version of this argument in Specificity Beats Sophistication: Why Generic AI Pages Are About To Lose The Search Results: the pages that win carry specifics nobody else has. A keyword export contains no specifics. It is the opposite of a moat. It is a shared driveway.
The report you already pay for
Your keywords are what you bid on. Your search terms are what people actually typed. Match types guarantee a gap between the two — phrase match stretches, broad match wanders, and Google's matching has only gotten more aggressive about it. That gap is not a problem. That gap is the editorial calendar.
The search terms report carries three columns no research tool can sell you:
- The exact string. Typos, word order, weird modifiers and all. The customer's phrasing, not the industry's phrasing about the customer.
- The price. You know to the cent what that intent costs on the open market, because you paid it. Sort by cost and the calendar doubles as a list of the queries most expensive to keep renting.
- The conversion. Not "search volume," which is an estimate of curiosity. A record of whether the person who typed that string bought from you.
There is one more thing the tools structurally cannot do. Google has said for years that about 15% of the queries it sees each day are ones it has never seen before. Keyword tools are built from history, so they show "no volume" for next year's language. The report hands you the new phrasing first — in your niche, attached to money — while every tool still insists the topic doesn't exist.
The pipeline
The mechanics are boring, which is the point. Boring is what automates.
- Export the report. Google Ads moves the menu around; today it lives under Insights and reports → Search terms. Schedule a CSV export, weekly, to a Drive folder or an inbox. Pull 90 days, not 7 — you want sample size, and conversions lag.
- Hand the file to Claude with your publishing MCP server connected. The whole job is one prompt.
You are connected to the htmlpub MCP server. Work through this list in order:
1. Read search-terms-last-90-days.csv from the exports folder.
2. Discard terms that are navigational (our brand, login, pricing),
competitor names we won't write about, and obvious broad-match noise.
3. Cluster what remains by the question the searcher is asking.
4. For each cluster, sum clicks, cost, and conversions.
5. Pull the list of live posts from the site and drop any cluster we
already cover.
6. Take the five highest-converting clusters left and draft one post each.
The title uses the searchers' own phrasing. The answer lands in the
first hundred words. Include the number, price, or name the query
implies they wanted.
7. Publish all five as drafts tagged "search-terms" and print the URLs.
- Do the human pass. This is step 8 and it belongs to you. Twenty minutes: cut the draft that's thin, fix the claim that's wrong, approve the three that are right. Then publish — the drafts are already sitting in the tool, because that's the part The Publish Button Your AI Was Missing covers.
Notice what step 6 is doing. The title mirrors the query word for word. Paid teams call this message match when the ad and the landing page agree. This is the organic version of the same law: the searcher sees her own sentence in the results and clicks it.
What the terms turn into
A sanitized slice of a report looks something like this:
| The search term, as typed | Clicks | Conv. | The post it becomes |
|---|---|---|---|
| kajabi alternative for one course | 84 | 6 | "Kajabi Alternatives When You Only Sell One Course" |
| how much should a sales page cost | 61 | 4 | "What a Sales Page Should Cost" |
| take payment on a one page site | 47 | 3 | "How to Take a Payment on a One-Page Site" |
| sales page examples for life coaches | 112 | 7 | "Sales Page Examples From Real Coaches, Annotated" |
Read the second row again. No keyword tool on earth has that row. Volume for "how much should a sales page cost" rounds to zero in every estimator, because estimators smooth the long tail into nothing. But you paid for 61 clicks from people asking it, and four of them converted. That is not a keyword. That is a receipt.
The rule that falls out: a converting query with no organic page is a lease you pay Google forever. The post is how you buy the building. You already know the exact phrasing, the market price of the click, and the close rate of the intent. Keyword tools give you one of those three, as an estimate.
What the report tells you not to write
The same file works in reverse. The terms you keep excluding as negatives are a record of how the market misunderstands you. If "free" modifiers keep eating budget, that is either a post that converts the free-seeker or a permanent negative — but it is a decision, and now it's a visible one.
Two filters stay manual. Competitor-name terms are tempting and usually a trap; only write the comparison you can write honestly, or the page reads as an ad and dies. And nothing in the pipeline autopublishes. Claude drafts, the MCP server stages, you decide. The moment a pipeline publishes without a reader, the report stops being an advantage and starts being a liability with your name on it.
One caveat worth keeping: since 2020, Google hides low-volume queries behind a privacy threshold, so the report is not the complete truth. It doesn't need to be. The terms that clear the threshold are the ones with scale, and those are the ones worth writing for.
The calendar that refills itself
Run the loop every Monday. It never empties, because match types keep drifting and roughly 15% of tomorrow's queries do not exist yet. The calendar refills itself from the traffic you are already buying.
It also has a sibling. The same account feeds the paid-side version of this pipeline — 40 Ad Groups, 3 Landing Pages: A Claude + MCP Pipeline for Real Message Match — where ad copy goes in and a brand-locked landing page comes out. One pipeline turns the account's inputs into pages. This one turns the account's exhaust into posts. The ad spend buys the data either way; the only question is whether you read it.
The keyword tool asks what the world might want. The search terms report records what your buyers wanted on Tuesday, what it cost you, and whether they paid. One of those is a guess. The other one is a receipt.