Stop topic chasing. Start winning the head-to-head.

Here is the pattern I keep walking into. A company runs a disciplined content program: keyword research, a publishing calendar, posts on every question the tools surface. The dashboards are green. And when a real buyer asks ChatGPT or Perplexity who to hire in their category, the answer recommends a competitor. In April 2026, a study of 150 SaaS companies found that 44% of the brands ranking in Google's top ten got zero mentions from ChatGPT on the same keywords. Page one, and absent from the answer.

I have a name for the machine that produces this outcome. I call it topic chasing, and naming it matters, because the tactics inside it are good enough to hide the problem.

Content is Code is my methodology, and the opening piece of this series argues it in full. The short version: query the AI platforms with real buying questions, decode why the winning answer wins, engineer that evidence on your own domain, and verify by rerunning the same questions. This article is about the first honest step: recognizing that the playbook most companies already run cannot see any of it.

Topic chasing, defined

Topic chasing is the standard playbook run without a recommendation scoreboard. It has three moving parts. Pull keyword demand out of a tool and rank the list by volume. Harvest the questions people ask around those keywords. Study what competitors publish, increasingly by feeding their posts to an AI model, and produce your own version. Ship it on a calendar, watch rankings and traffic, repeat.

Read that list again and notice something: every tactic in it is legitimate. Keyword demand is real evidence of what buyers ask. Question harvesting is real audience research. Studying the competition is as old as competition. I concede all of it, honestly, because the argument I am about to make does not need a straw man.

What topic chasing gets right

The coverage this playbook produces is real fuel for AI visibility, and the evidence says so. Google states plainly that its AI features, including AI Overviews and AI Mode, are built on the same core search ranking and quality systems as classic search, and that no special AI file or markup is required. Work that improves your standing in those systems feeds Google's AI surfaces directly.

The correlation evidence points the same direction. An Ahrefs study of 75,000 brands from May 2025 found branded web mentions were the strongest factor they measured against AI Overview visibility, at a 0.664 correlation, three times stronger than backlinks. The authors stress that every factor they tested was moderate at best and that correlation proves no causation. The brands in the top quartile for mentions averaged more than ten times the AI Overview presence of the next quartile. Broad, consistent coverage of your name and your subject matter is associated with showing up. So no, the program is never wasted. It is producing inputs.

What topic chasing cannot tell you

Now the ceiling. On AI assistants, ranking and being cited are different events. In August 2025, Ahrefs ran 15,000 long-tail queries through Google and Bing, then asked ChatGPT, Gemini, Copilot, and Perplexity the same questions. On average, only 12% of the links the assistants cited ranked in Google's top ten for the same prompt. Eighty percent ranked nowhere in Google's top 100. The study covered long-tail informational questions, where overlap runs lowest, so hold the exact digits loosely. The direction still stands. Google's own AI Overviews sat at the other pole, drawing 76% of citations from top-ten pages. The assistants retrieve differently by design: they fan one question out into many related searches and fuse the results, a mechanism the Ahrefs study documents and Google's own guidance names as query fan-out.

The SaaS study completes the picture from the brand side: 44% of page-one brands got zero ChatGPT mentions, and 81% of the brands ChatGPT recommended did not rank in the top ten at all. One snapshot, one platform, one country, and the direction is unmistakable. You can win the topic and lose the recommendation. Nothing in the topic-chasing loop would ever tell you.

Inputs and the scoreboard, side by side

The difference is what each one measures, and a side-by-side makes it hard to unsee.

Inputs against outputs. Neither column replaces the other: the left side is how you produce, the right side is how you find out whether production wins. A content program measured only by topic-chasing numbers is a factory with no view of its own loading dock.

The head-to-head is the finish line the program was missing

The fix is additive. Keep the research, keep the calendar, and give the program a scoreboard: a locked series of real buying questions, locked meaning the same questions repeated unchanged on a cadence, run across the major platforms, recording who was recommended, who was named, and the stated reasons every time. That win rate against the competitors AI actually names is the top layer of measurement, and the lens this whole series runs on.

With the scoreboard in place, the tactics get sharper on their own. Keyword demand still tells you what buyers ask, and now the losing comparisons tell you what to publish about: the evidence the winning answers cite that your domain does not carry. That is the loop in the key card above, run as an operating rhythm. And the needle can move. Soapbox Bulletin, a client I can name, went from a 6.2% to a 36.2% AI recommendation win rate in eleven weeks working this way. Win rate here means the share of locked head-to-head questions the platforms answered in the client's favor. That is directional evidence rather than a controlled study, and I report it that way every time. The full write-up, limits included, lives on the case studies page.

Change the finish line this week

Three moves, none of which require pausing the content program. First, run one head-to-head pass: the major platforms, a handful of real buying questions, three columns of notes. A real buying question sounds like a buyer, so ask it the way one would: “best commercial HVAC contractor for a hospital retrofit in Austin,” never “HVAC services.” Second, put the competitor names AI gave you next to the list in your deck and log the differences. Third, pick the loss that stings most and trace the winning answer's evidence back to its sources. That trace is next quarter's content brief, and the scoreboard produced it.

Topic chasing feels productive because it is production. The head-to-head is the part of the job that tells you whether production is winning anything. First measure who is winning. Then change it: query, decode, engineer, verify.

Questions worth asking next

Sources

1. Google Search Central, "AI features and your website". https://developers.google.com/search/docs/fundamentals/ai-optimization-guide
2. Louise Linehan and Xibeijia Guan (Ahrefs), "An Analysis of AI Overview Brand Visibility Factors (75K Brands Studied)," May 26, 2025. https://ahrefs.com/blog/ai-overview-brand-correlation/
3. Louise Linehan and Xibeijia Guan (Ahrefs), "Only 12% of AI Cited URLs Rank in Google's Top 10 for the Original Prompt," August 11, 2025. https://ahrefs.com/blog/ai-search-overlap/
4. Matt Shirley (EMGI Group), "The SaaS AI Citation Gap Report 2026," April 2026. https://emgigroup.com/blog/saas-ai-citation-gap-report/