Short answer
Run one content plan, not two. Build it from the questions your buyers actually ask, give each question exactly one page that owns it, and open every page with an answer that can be lifted out and quoted on its own. Size the calendar to your review capacity rather than your writing capacity — the bottleneck in a small company is always the person who can tell whether the article is true. Then measure two separate things: position, from Search Console, and mention, from asking the questions yourself and recording who gets named.
One plan, two audiences
The most expensive mistake in content strategy right now is running a GEO program alongside an SEO program. It doubles the meetings, splits the budget, and produces two half-resourced efforts whose outputs contradict each other.
One plan is enough because the two audiences want the same thing at the source. A search engine ranks a page it can crawl, read and trust. A generative engine quotes a passage it can fetch, extract and trust. The overlap is nearly total up to the point of extraction. What differs is the shape of the page, and shape is a brief-level decision — not a program-level one.
We covered the underlying distinction in what generative engine optimization actually is. This article is the operational layer above it: what you decide, in what order, before anyone writes a word.
A useful sanity check before you approve any content plan: could you hand it to a new employee and have them explain, for each planned page, which single question it answers and who is meant to be searching that? If the plan only contains topics and volumes, it has not been made yet.
Start from questions, not keyword volumes
Keyword tools report how often a string is typed into a search box. That was a good enough proxy when the search box was the only entrance. It is a poor proxy now, for two reasons that compound.
- Conversational queries are longer and more varied. People type into an assistant the way they would speak to a consultant. Those phrasings are individually rare, so a volume tool rounds them to zero and your plan never sees them.
- Volume says nothing about intent to buy. The question that precedes a purchase decision — "is this worth doing for a company our size", "what does it actually cost to run" — is asked far less often than the definitional query, and is worth far more.
The replacement input is unglamorous and free. Collect the questions your team already answers: the ones in sales calls, the ones in support tickets, the ones a prospect asks in the last ten minutes before signing. Those are questions with money attached, phrased the way buyers phrase them.
Keep the keyword tools, but demote them. Their remaining job is to tell you which of your real questions also happens to have search volume, so you can sequence those first. They are a sorting input, not a sourcing input.
Map the questions before you write anything
Once you have thirty to fifty real questions, the strategy step is grouping them — and the grouping rule is stricter than most teams apply: one question, one page, one owner.
Two pages competing for the same question is not redundancy insurance. It splits whatever authority the topic accumulates across two URLs, and it leaves a retrieval engine choosing between two of your own pages, which it may resolve by choosing neither. If two planned articles answer the same question, one of them is a section of the other.
A structure that holds up in practice has three tiers:
- One pillar per commercial topic. The broad page a stranger lands on. It defines the terms, states your position and links out. It rarely wins a citation on its own, and that is fine — its job is to hold the topic together.
- Specific answers underneath it. One page per real question, each self-contained enough to be quoted without the pillar. This is where citations actually come from.
- Comparisons and decisions. The pages people read when they have narrowed to two options. Our own tool comparisons sit here, and they attract a different, later-stage reader than either tier above.
Link the tiers to each other with plain descriptive text. A link labeled "learn more" tells a machine nothing about what sits on the other side; a link labeled with the question the target page answers tells it everything.
The brief that serves both readers
Everything above is planning. The brief is where the plan becomes a page that can rank and be quoted, and it needs only six lines.
- The one question this page owns, written as a question, in the buyer's words.
- The answer, in under sixty words, written before the article. If it cannot be written, the question is not yet understood well enough to publish on.
- Who this is for, specifically enough to exclude someone. "Operations managers at 20 to 80 person manufacturers" beats "businesses".
- What we can prove, and where the proof comes from. A figure with no traceable source does not go in.
- What we will not claim. The caveats, the cases where this does not apply, the honest limits. This section is what stops an article aging into a liability.
- Who reviews it before publication, by name.
The sixty-word answer goes at the top of the finished page, before the context and the atmosphere. That single habit does more for citation than any schema markup, because it is the passage a retrieval system can lift intact.
Cadence, and why review capacity is the real limit
Publishing cadence is usually planned against writing capacity. That is the wrong constraint. Drafting is the cheap part now; the scarce resource in a small company is the person with enough domain knowledge to say whether a paragraph is true.
So size the calendar to that person's hours. If your operations lead can review two articles a month without it eating their week, your cadence is two articles a month. Publishing six that nobody qualified read is not three times the output — it is three times the exposure to being quoted saying something wrong, on a surface where you cannot see the quote happening.
The asymmetry is worth stating plainly. An unpublished article costs you an opportunity. A published article containing a false claim costs you the credibility of every other page on the domain, and you will not find out from an analytics dashboard — you will find out when a prospect repeats it back to you in a meeting.
Consistency beats volume for a second reason: a topic covered thoroughly over a year reads as expertise, while a burst of twelve articles in one month followed by silence reads as a campaign. Google's guidance on creating helpful, reliable, people-first content asks whether content is produced mainly to attract traffic — and a monthly burst pattern is one of the few things about a calendar that is visible from the outside.
Refresh, retire, or write new
After the first year, most of the value in a content program sits in deciding what to do with what you already published. Three outcomes, and a test for each.
- Refresh when the question is still the right one and only the answer has moved. Rewrite the substance, update the modification date honestly, and keep the URL. You keep every signal that URL has accumulated.
- Retire when the question is no longer asked, or when the page exists only because a keyword tool suggested it. Redirect it to the page that covers the nearest live question. A thin page that nobody reads still gets crawled, and still forms part of how your site is assessed.
- Write new only when the question is genuinely distinct from every question already covered. Say it out loud next to the existing titles; if it sounds like a rephrasing, it is one.
One discipline about dates, since it is easy to get wrong in the direction that feels productive: update the modification date when the content changed, and not otherwise. This is not a matter of taste. Google's self-assessment list for search-engine-first content asks it directly — "Are you changing the date of pages to make them seem fresh when the content has not substantially changed?" — and answering yes to it is described as a warning sign to reevaluate how you produce content.
Measuring a plan that has two outcomes
One plan, two measurements — and they are not interchangeable.
Position is the easy half. Search Console reports impressions, clicks and average position, and Google's own documentation on its AI features states that sites appearing in AI Overviews and AI Mode are included in overall search traffic, reported inside the Performance report under the "Web" search type — not broken out separately. Treat those numbers as your search picture rather than your AI picture, and read what actually makes a page eligible for AI Overviews before drawing conclusions from a drop.
Mention is the half you have to build yourself, and it is a manual monthly baseline rather than a dashboard. We set out the method — which questions, how to run them, and why a single measurement is unreliable — in how to get cited by ChatGPT and Perplexity. The strategy decision here is narrower: the questions you measure must be the same questions your map assigned pages to, or the number tells you nothing about the plan.
Both measurements are lagging. The leading indicator worth watching in month one is simpler: what percentage of published pages have a sixty-word answer in the first screen, and a named reviewer. That is entirely within your control, and it is the input the other two numbers depend on.
Frequently asked questions
Do I need a separate content plan for AI search?
No. A second plan produces two thin programs instead of one good one, and most of the work is shared anyway: research the real question, answer it plainly, publish somewhere a crawler can reach. What changes is the brief for each page, not the calendar. Give every page one question it owns and one extractable answer near the top, and the same page competes in both places.
How many articles should we publish per month?
Fewer than most agencies sell, and more consistently than most companies manage. For a 10 to 100 person business, two genuinely useful pages a month that someone in the company actually reviewed beats eight generated ones. The constraint is not writing capacity, it is review capacity: an article nobody with domain knowledge checked is the one that gets quoted back at you incorrectly.
Should we update old articles or write new ones?
Update first when the page already covers the right question and has simply gone stale, because you keep whatever history the URL has accumulated. Write new when the question is genuinely different. The failure case is publishing a second article on the same question, which splits the signal between two URLs and leaves a retrieval engine to guess which one you meant.
Does publishing more content improve AI citations?
Not on its own. Retrieval engines fetch a passage that answers the question asked, so a hundred pages that circle a topic lose to one page that states the answer. Volume helps only in the narrow sense that more distinct questions covered means more distinct queries you are eligible for. Publishing the same idea repeatedly under different titles adds nothing a model can use.
If the plan is clear and the bottleneck is doing it, that is the part we take on — the question map, the briefs, the review loop and the monthly citation baseline. Book a call and we will start from the questions your sales team is already answering.
