How Lean Teams Should Decide What to Write Next: A Prioritization Framework

You have twenty good ideas and time for two. That is not a strategy problem. That is a ranking problem, and it is the one that quietly eats small content teams alive. This guide gives you a content prioritization framework you can run in an afternoon, so deciding what to write next stops being a debate and starts being a decision you can defend.

Here is the short version. Score every candidate topic from 1 to 5 on reach, impact, AI-citation opportunity, and confidence. Estimate effort on the same 1 to 5 scale. Then calculate:

Priority score = ((Reach × Impact × AI-citation opportunity × Confidence) ÷ Effort) × 20

That formula borrows its bones from RICE, the product prioritization method built on reach, impact, confidence, and effort. The one change that matters for content in 2026: AI-citation opportunity gets its own column instead of hiding inside a general guess about impact.

One thing to set expectations. The scores below are an operating convention, not an industry benchmark. They rank your options under uncertainty. They do not predict traffic, citations, or revenue, and you should recalibrate the scales once you have real outcomes to compare them against. What they do fix is the part that hurts most: lean team content decisions made by whoever argued hardest in the meeting.

Ready? Let's go step by step.

Step 1: Set your objective and your capacity for this round

Start by naming two things: what you are trying to achieve this quarter, and how much work you can actually absorb.

Write down one primary objective. Something like "create qualified organic discovery," or "support the consideration stage," or "close the visibility gap against our two closest competitors." Then define what "next" means for you. The next single piece? The next three? The next forty available team-hours?

You are done with this step when every person scoring can answer three questions: who are we trying to help, what outcome matters, and how much capacity do we have. If your objective is vague enough that it would never change a ranking, it is not an objective yet.

Where teams go wrong: mixing objectives without saying so out loud. A broad awareness guide, a high-intent comparison, and a technical decision page can all be excellent. They are not doing the same job, and scoring them as if they were produces a mush of a number nobody trusts.

Pro tip: run one primary model. If you genuinely need a second view, run a separate scenario, for example a pipeline-weighted ranking next to an AI-gap-weighted one. Do not average two incompatible objectives into one opaque score.

Step 2: Build a candidate list from evidence, not from a brainstorm

A brainstorm gives you titles. You need candidates with reasons attached.

Pull from five places:

1. Questions that keep coming up in sales, customer success, support, and community threads.
2. Search Console queries and pages with relevant impressions, clicks, and position data.
3. Tracked AI prompts where your brand is absent, under-mentioned, or not cited.
4. Competitor pages covering a buyer need your site handles badly or not at all.
5. Thin or missing areas on your own site, plus any overlap that might need consolidating instead of a new page.

Write each one as a proposed page angle, not a keyword. Include the audience, the buyer stage, the problem, the answer you intend to give, the evidence behind it, and any page it might overlap with.

You are done when each row is specific enough that two colleagues would read it and picture the same article. "AI search" is not a candidate. "How a B2B marketing lead can audit which competitor pages AI engines cite for buyer questions" is.

Where teams go wrong: copying competitor headlines, treating every keyword variation as its own article, and ending up with forty near-duplicates. Google's people-first guidance warns against producing lots of content on many topics in the hope that some of it performs. Group candidates by the user need behind them before you score anything.

This is one of the steps where tooling genuinely saves you hours. DeepSmith's Content Map crawls your site and your competitors' sites into one shared topic taxonomy, classifies every page by granular topic and funnel stage, and surfaces two things you would otherwise assemble by hand: coverage gaps where a competitor publishes more than you, and untapped topics where a competitor publishes and you have nothing at all. Opportunity Agents go a step further and return ideas with the specific data point that justifies each one attached. That gets you an evidence-backed list. It does not decide your business priorities, and it should not.

Step 3: Give every candidate an audience, an intent, and a business job

A keyword is not a strategy. You cannot prioritize content topics you have not defined, and five labels turn a vague candidate into something you can actually score.

For each one, record:

• Audience: the persona or role you are writing for.
• Buyer stage: Awareness, Consideration, or Decision.
• Intent: learn, evaluate, compare, solve, implement, or choose.
• Business job: educate the market, create qualified discovery, support an evaluation, answer an objection, or strengthen a strategic topic.
• Success signal: the one measure you will look at, whether that is relevant impressions, qualified clicks, sales usage, mention rate, or citation rate.

Then write a single sentence: "This page should help [audience] do [job] so that [business outcome] becomes more possible."

You are done when that sentence names a real person and a real task, and the page would still be useful to someone who arrived from a conversation rather than a search result.

Where teams go wrong: confusing buyer stage with importance. A big awareness topic can matter a lot. It should not automatically outrank a smaller, high-intent topic just because the volume number is prettier.

Google's own people-first self-checks are a good sanity test here. Would your intended audience find this useful? Does it show first-hand expertise? Will the reader learn enough to actually finish the task? Does it offer original information, analysis, or genuinely additional value? If you cannot answer yes, the problem is upstream of your score.

Step 4: Check what you already cover before you score the upside

This step saves more wasted work than any other, and almost everyone skips it.

For each candidate, look at your own site and record:

• Existing pages serving the same or an adjacent intent.
• Whether coverage is absent, thin, outdated, top-heavy, or already strong.
• Whether this needs a new page or a better version of an existing one.
• Internal links that could connect the new piece to what you already have.
• Which funnel stage, if any, is missing.

Then compare against competitors, and keep one distinction sharp: "a competitor has a page" is very different from "a competitor has a good, differentiated page." A gap is a lead. It is not proof of demand.

You are done when every candidate carries a disposition: write it new, improve an existing page, merge it with another candidate, or reject it. And when no two high-ranked candidates are chasing the same underlying intent without a reason.

Where teams go wrong: rewarding the biggest visible gap, then publishing a second page that splits the same intent with a page you already have. More URLs does not equal more authority. Google says internal links help it find new pages and act as a relevance signal, which supports clear structure and useful cross-referencing, not page volume for its own sake.

Content Map reports per-topic page counts and funnel distribution, so you can see at a glance whether a topic is thin, top-heavy, or missing its decision-stage pages entirely. It re-checks sitemaps every 24 hours, so newly published pages fold in without a re-import. Treat that as a workflow convenience that keeps your coverage picture honest, not as any kind of performance promise.

Step 5: Estimate reach with the strongest proxy you have

Reach is how many relevant people, searches, accounts, or buyer situations this page could affect in a defined window. Not raw search volume. Relevant reach.

Pick one primary proxy per candidate and write down its source and its observation window. Prefer first-party evidence, in this order:

1. Search Console impressions and clicks for related queries or pages.
2. Tracked AI prompts, and how many prompt groups or engines are affected.
3. Recurring customer, sales, or support demand.
4. Keyword or market-demand estimates, clearly labeled as estimates.
5. A strategic audience-size hypothesis, when you have no direct demand data at all.

Score it 1 to 5. A 1 means no evidence of a meaningful audience. A 3 means clear recurring demand from Search Console, customer conversations, or a relevant prompt set. A 5 means broad, strategically important demand across several audience or prompt groups.

You are done when the reach score has a written reason, a source, and a time window beside it.

Where teams go wrong: treating impressions as attention, clicks as pipeline, or keyword volume as qualified demand. Search Console defines impressions as how many times your site appeared in results, and CTR as clicks divided by impressions. Both are directional. Neither proves business value on its own. Search results also vary by time, place, device, and recent history, so one manual search is not a market-size measurement.

Step 6: Score the AI-citation opportunity from gaps you actually observed

This is the column most content topic scoring models are missing, and it is the reason this content prioritization framework has five components rather than four.

First, keep two words separate. A mention is when an AI answer names your brand. A citation is when the answer links to one of your pages as a source. You can be mentioned without being cited, cited without being recommended, or missing from both. Collapsing them into one number hides the gap you are trying to find.

Build an AI-gap evidence row for each candidate. Check the buyer prompts you actually track, on the engines you actually track, and record:

• The exact prompt, or a stable prompt family.
• The engine and the date you checked.
• Whether your brand was mentioned.
• Whether one of your pages was cited.
• Which competitor or other source was cited instead.
• Whether you already have a page that could satisfy that prompt.
• Whether the gap is a missing topic, a weak page, unclear positioning, or just noise.

Then score 1 to 5:

• 1: no observed gap. Your page is already cited consistently, or the topic has no tracked relevance.
• 2: a weak or isolated gap, or a low-value prompt with no obvious page-level response.
• 3: you are absent or under-cited for a relevant prompt group, and a credible page could address it.
• 4: repeated absence or competitor citation across important prompts, with a clear content gap or a weak existing page.
• 5: high-value tracked prompts repeatedly cite competitors or omit you, the topic is strategically important, and you have a credible original angle.

You are done when the AI score traces back to prompt-level evidence and you can state exactly which gap the page would close.

Common mistake: A competitor citation is not a content brief and not a guarantee. Record the prompt, engine, date, cited page, and the missing user need before you award a high AI-opportunity score.

Do not hand out a 4 or 5 because "AI likes concise answers." Google's current guidance on AI features says the same foundational SEO best practices apply, with no additional requirements and no special optimization needed. It also says a page must be indexed and eligible to appear in Search with a snippet to be eligible as a supporting link, and that meeting best practices never guarantees crawling, indexing, or serving.

Collecting this evidence by hand across several engines is where most lean teams give up. DeepSmith's AI Visibility reports mention rate, citation rate, share of voice, sentiment, and visibility trend, broken down per platform, with a competitor leaderboard and the sources AI cites most often. The Pages view shows which of your pages AI actually cites and which prompts drive those citations, and the competitor view shows which of their pages win for your tracked prompts. That is exactly the input this column needs. DeepSmith reports the visibility data. It does not control or guarantee rankings, citations, traffic, or revenue.

Step 7: Score impact, confidence, and effort, then run the formula

Three components left. Take them one at a time.

Impact, 1 to 5. The expected business value if this page reaches the right people and does its job. Weigh product relevance, buyer stage, commercial intent, whether it shows your expertise, whether sales or support can use it in a real conversation, and whether it fills a real hole in a coherent topic area. Require a one-sentence "why it matters" note. If nobody can name the audience, the problem, and the outcome, impact caps at 2.

Confidence, 1 to 5. How strong the evidence is behind your estimates and your angle. A 1 is mostly intuition. A 3 is at least two independent signals, such as customer questions plus competitor coverage, or Search Console demand plus a tracked AI gap. A 5 is strong first-party evidence, clear demand data, a documented gap, a defined audience, and a scope you have validated. Keep confidence separate from impact. A very important idea with low confidence means validate it, not discard it.

Effort, 1 to 5. Total team-hours to produce a credible, publishable page. Not drafting time. Include research, subject-matter input, review, fact checking, design or data work, approvals, and coordination.

• 1: roughly 1 to 2 hours. A narrow update or simple explainer.
• 2: roughly 3 to 5 hours. A focused piece with evidence close to hand.
• 3: roughly 6 to 10 hours. A normal how-to, comparison, or substantial guide.
• 4: roughly 11 to 16 hours. A deep guide, original analysis, or expert interviews.
• 5: more than 16 hours, or a real cross-functional dependency.

Those bands are a starting convention. Swap them for your own observed median once you have enough finished pieces to know it. If a candidate needs engineering, proprietary data, or an expert who is never free, write that dependency down instead of pretending it is a 3.

Now put it all in one sheet:

Calculate the score exactly as written, and keep the raw component values visible. That last part matters more than it sounds. When a stakeholder disagrees, you want the argument to be about the reach number, not about a mysterious total.

Here is the arithmetic on three made-up candidates, purely to show how the formula behaves:

• Candidate A: Reach 4, Impact 5, AI opportunity 5, Confidence 4, Effort 3. Score = ((4 × 5 × 5 × 4) ÷ 3) × 20 = 2,666.67.
• Candidate B: Reach 5, Impact 3, AI opportunity 2, Confidence 5, Effort 2. Score = ((5 × 3 × 2 × 5) ÷ 2) × 20 = 1,500.
• Candidate C: Reach 3, Impact 4, AI opportunity 4, Confidence 3, Effort 1. Score = ((3 × 4 × 4 × 3) ÷ 1) × 20 = 1,440.

Those are invented values, not results. What they show is the shape of the model. Multiplication means one weak component drags the whole thing down, which is the point. A huge audience with no business value should not win. And because effort sits in the denominator, a small easy piece stays competitive.

Where teams go wrong: giving everything a 4 or 5, treating effort as an afterthought, or scoring a topic high because a senior person likes the title. The model only helps when the differences mean something.

Step 8: Apply the decision bands, break ties, and commit your shortlist

Sort descending, then read the score against these bands:

• 2,000 or higher: a strong candidate for the next writing slot, assuming it clears your filters.
• 1,000 to 1,999: viable. Pick it when it serves the current objective or fills a gap that matters strategically.
• Below 1,000: validate, narrow, merge, improve an existing page, or reject. Do not write it by default.

Those cutoffs are recommended starting points for this formula, not validated thresholds. Recalibrate once you have enough completed decisions to compare scores against what actually happened.

Before you sort, run the hard filters. These are pass or fail, not scores:

1. Audience fit: reject anything with no identifiable reader.
2. Business fit: reject anything you cannot connect to a current objective or customer problem.
3. Evidence floor: require at least one concrete evidence item, and two independent signals for anything on the shortlist.
4. Quality floor: reject topics where you cannot bring original experience, analysis, data, expert input, or a genuinely useful synthesis.
5. Overlap check: merge or revise anything competing with an existing page for the same intent.
6. Feasibility check: flag anything blocked by missing expertise, data, approvals, or product claims.

A high score with no audience is still a no. Filters are not negotiable and scores are.

For ties, work down this order: higher impact, then higher AI-citation opportunity where the gap is prompt-level and repeated, then lower effort, then higher confidence, then stronger originality, then better fit with an existing topic area.

If two candidates are still level, run the smallest validation that could change the answer. Ask sales which question comes up more. Re-check the prompt evidence. Get a fifteen-minute subject-matter review. Narrow one of the angles. Validation is not production, and it should not become a way to avoid deciding.

You are done when you have a short ranked list, a written reason the top candidate won, and a recorded "not now" reason for the runners-up. That last one is worth the two minutes it takes. Next month, you will not have to re-litigate the same debate from scratch.

Where teams go wrong: treating the ranking as permanent, or using the score to dodge editorial judgment. The model makes trade-offs visible. It does not make the final call for you.

What to do next

Run this on your current backlog once. Not perfectly, just once. Score five candidates, calculate, and see whether the ranking matches your gut. Most lean team content decisions get easier the moment the disagreement has a column name. Where it does not, the disagreement will point at exactly which component you are estimating badly, and that is genuinely useful information.

Then leave it alone for a cycle. When those pieces have real outcomes, compare them with the scores you gave them and adjust the scales. Recalibrating beats adding more components every time.

The hardest input to gather by hand is the AI-citation opportunity column, because it means checking buyer prompts across several engines and logging what you find. If you want that evidence collected for you, alongside a coverage map of your site and your competitors', start a free DeepSmith trial and see what your gaps look like with real data.

You do not need a bigger backlog. You need a smaller, better-argued shortlist. This gets you there.