Prioritize 5 Articles to Write This Week Using GEO Data

Keyword research tells you what people type into Google. It says nothing about which brands an AI engine names when someone asks for the best option in your category. This workflow starts from the monitored questions where competitors appear and you do not, and ranks the briefs on the score Qwairy already computes. Prompt signals appear here as a filter only; the portfolio playbook is where you learn to read them.

The GEO problem it solves

Search ranking and AI citation are two different measurements. A page can rank first on Google and never be quoted by an assistant, which pulls from the sources and framings it already trusts for that question.

The ranking is the second problem. Editorial planning usually ends with a list ordered by someone's feel for impact, which is unarguable. The Action Center already scores its recommendations from 0 to 100, so the last step is a sort.

How the workflow breaks down

1. Surface the gaps. get_content_opportunities returns the questions where direct competitors are named and you are not, plus those answered at least twice where you appear in under 30% of answers. Both cap at 10 rows, so a thin result means a narrow prompt set, not a saturated market.

2. Filter by strategic signal. get_prompt_signals sorts your prompts into attack, defend, monitor and ignore, and returns the top attack and defend prompts with share of voice and a priority score, adding an openness value on the attack list only. Any of the four counts can be zero; when both shortlists are empty, the step 1 questions carry the workflow forward.

3. Widen each question. get_query_fan_out returns the web-search queries observed inside monitored answers, each with the prompts it came from and your presence rate. A query tied to several prompts covers more ground per article. Answers that never trigger a web search produce nothing.

4. Read what wins today. get_prompt_answers, then get_answer_details, exposes the generated text: who is named, in what order, citing which sources.

5. Rank on the scored queue. get_actions returns the live queue, suggested plus pending plus in-progress by default, each row carrying a 0 to 100 priority score, an effort label and a checklist. Filtered to CONTENT_GAP and CONTENT_OPTIMIZE, it is your candidate list, already scored.

What a run actually returns

A real 30-day window returned the cap on both halves: 10 missed questions and 10 low-visibility ones. Every low-visibility row sat at 0 percent across 13 to 16 answers each. Prompt signals on the same set returned 0 attack and 0 defend out of 155 prompts.

Both fallbacks fire at once, which is the case worth planning for. Ten missed questions at the cap means the real number is higher and the list truncated, so competitor count is the only honest sort. And a low-visibility row at 0 percent over 16 answers is not a weak position to defend, it is an absence: the same work as a missed question under a different heading.

What to do with the output

Record each target prompt's current state before anything ships. You publish against the prompts, not the article, and that snapshot makes the after readable.

If both type filters return nothing there is no scored queue to sort, so fall back to the step 1 questions ordered by distinct competitor count, and say the ordering is yours.

Otherwise sort the returned actions by priority score, highest first, and fill the week from the top. The score is tier anchored: 70 and above is the critical band, 40 to 69 the high band, below 40 the standard band. Stop at 40. If only three rows clear it, write three articles rather than promote a standard-band item. On an equal score, take the LOW effort row first.

The Complete Prompt

I want five article candidates to write this week, ranked on my GEO data. List the brands I monitor and, if there is more than one, ask which to use before pulling anything.

Walk me through this step by step, explaining what each step reveals:

1. Pull content opportunities for the last 30 days. Show the missed and the low-visibility questions, ordering the missed ones by distinct competitor count.
2. Run prompt signals with limit 10. Show the four category counts and the top attack and defend prompts. If both shortlists come back empty, say so, skip the signal filter and carry the step 1 questions forward.
3. Run query fan-out for the last 30 days. Map the queries back to the selected prompt IDs using the prompts field on each row. If nothing is returned, say so and move on rather than inventing a tree.
4. For up to five selected prompt IDs, pull the prompt answers, then the answer details on the most relevant ones. Report the competitors named, their order and the sources cited.
5. Pull actions with type CONTENT_GAP, then again with type CONTENT_OPTIMIZE. Discard any row that is not an article to write, sort the rest by priority score highest first, and keep the top five with title, score, effort, checklist, target prompts, competitor context, and the facts a writer must verify.

Apply these ranking rules literally. Keep only rows scoring 40 or more, return fewer than five candidates if fewer clear it, and tell me how many you dropped. Break ties by effort, LOW before MEDIUM before HIGH. If both calls return nothing, say the queue holds no content item, then rank the missed step 1 questions by distinct competitor count and the low-visibility ones by visibility rate ascending, labelling the ordering as your own. If step 1 was also empty, stop and say no gap showed this window.

Close by recording each target prompt's current visibility and share of voice as my before state.