Most teams do two of the three steps. They map the questions their buyers ask, and then they write. Map, then write. The step they skip is the one in the middle, and it is the one that decides whether the other two were worth doing.
You finish the mapping work and you have a list of fifty or sixty real buyer questions across Google, ChatGPT, Perplexity, and the forums. You can publish maybe six pieces a month. So the real question is not “what could we write.” It is “which six, in what order, and why those six and not the other fifty.” That is the Target step, and almost nobody does it on purpose. They sort by gut, or by search volume, or by whatever the loudest person in the room argued for on Monday.
Why the SEO reflex fails here
The instinct most people bring is the one that worked in Google for fifteen years: sort the list by search volume, chase the big numbers. For AI search that signal is close to useless.
A query can have low reported volume and still be the exact question a buyer types into ChatGPT the week before they choose a vendor. Volume tools barely see those queries, because they happen inside a chat window, not a search box. Meanwhile a high-volume informational term like “what is generative engine optimization” can be answered by the engine from memory, with no source pulled at all, which means no page you write this quarter is going to change the answer. Big number, zero leverage.
So the volume sort is not just weak. It actively points you at the wrong work.
The three filters I rank a query by
When I turn a map into a plan, every candidate query passes through three filters, roughly in this order.
Buying intent. How close is this question to a purchase decision? Category and comparison queries (“best issue tracking tool for software teams,” “Linear vs Jira,” “how do I track AI search traffic”) sit right next to the buying moment. Top-of-funnel “what is X” questions sit further away and convert worse even when you win them. I weight queries by proximity to the decision, not by how many people ask them.
Retrieval mode. Does the engine actually fetch live sources for this question, or does it answer from memory? This is the filter everyone misses, and it is the difference between a winnable race and a wall. If the engine retrieves live, a clear new page can show up within weeks. If it answers from training memory, the only way in is to have been prominent in the data for a long time already, and no single page fixes that fast. I covered the mechanism in how AI engines decide what to cite. For planning, the rule is simple: memory-answered queries get deprioritized for the near term, no matter how attractive they look.
Recoverable gap. This is the output of the competitor work. Where are you absent, or present but mislabeled, and the competitor sitting in your spot is thin enough to displace? A big recoverable gap on a query you can own beats a crowded query where the best you would do is sixth in a list the buyer stopped reading after the second name. The map tells you what buyers ask. The content gap analysis tells you which of those you can actually take.
A query that scores well on all three, high intent, live retrieval, recoverable gap, goes to the top. A query that fails any one of them drops, even if it is tempting.
Ranking is not the plan. The mix is.
Once the list is ranked, the second half of Target is choosing the shape of the month, not just the order. The split I run for clients is two flagship long-forms and four shorter pieces, every month.
The two flagships are the pieces built to get recommended by AI engines and earn links. They take the highest-intent, most winnable category questions and answer them more completely and more honestly than anything else on the page. These are the pieces that show up when a buyer asks an engine what the best option is.
The four shorter pieces take the specific comparison and troubleshooting queries. Tighter scope, faster to ship, structured so an engine can lift a clean answer straight off the page. They will not each carry the brand, but together they catch the long tail of exact questions a buyer asks on the way to a decision.
Two and four is not a magic ratio. It is a forcing function. It stops you from spending a whole month on one ambitious pillar that may not land, and it stops you from shipping six thin pieces that each move nothing. The discipline is the constraint.
What I got wrong first
For a while I sorted the map purely by gap size and went after the biggest holes first. It felt rigorous. It was not.
Half of the biggest “gaps” were high-volume head queries that the engines answered from memory. I burned a flagship on one of them, a broad category-definition piece, and watched it do nothing in AI answers for a full quarter, because the engine was never going to retrieve a source for that question in the first place. The gap was real. The query was unwinnable in the timeframe. I had skipped the retrieval-mode filter because the gap looked too good to pass up.
Now retrieval mode is a hard gate that runs before gap size, not after. A query the engine answers from memory does not enter the plan as a near-term flagship, no matter how empty the gap looks. That one reorder changed the hit rate more than any writing improvement did.
The plan is alive, not a list you file
A content plan for AI search is not a spreadsheet you fill in once. The engines change what they retrieve, competitors fill gaps, and you win some queries and free up budget for the next ones. So the plan gets rebuilt every cycle, against fresh measurement.
That is why measuring AI search visibility sits next to planning, not after it. You re-check where you stand, you see which targets you took and which moved, and you re-rank. A query you lost last month might have a thinner competitor this month. A query you won drops off the list and makes room.
Why this matters
Capture gives you the map. Investigate gives you the gaps. Target turns those into a finite, ranked, shippable plan, and Engineer builds each piece to get cited. Skip the middle and you get the most common failure I see: teams producing a lot of content and getting recommended by nothing, because every piece was chosen by instinct instead of by intent, retrieval, and gap.
The teams that win AI search are not the ones publishing the most. They are the ones who decided, on purpose, what not to publish.
If you want to see which queries in your category are actually winnable right now, that is what a paid audit produces, and the full framework is on the methodology page.