Prioritize GEO Budget Across Your Prompt Portfolio

During annual planning, SEO, paid, content and brand each file their own AI-related ask, and one at a time they all look reasonable. This workflow does not allocate the money: it is a duplicate-ask arbitration that finds the asks quietly buying the same monitored question twice and puts one name on each. It consumes the prompt-signals shortlists as an input; the portfolio playbook is where you learn how those shortlists are built and where they mislead.

The GEO problem it solves

Arbitration by channel is the habit: each ask judged against last year's line for its channel. In AI answers the unit of competition is the question, and two teams can target the same monitored prompt through an article and a review platform without either line noticing.

Comparing each ask against the current state of a prompt, who gets named and how open the answer still is, puts them all on one axis. That evidence settles which questions are contested, not what one is worth.

How the workflow breaks down

1. Pull the portfolio view. get_prompt_signals returns the attack, defend, monitor and ignore counts, aggregate openness, visibility rate and share of voice, then the top attack and defend prompts with a priority score on each. Monitor and ignore come back as counts only, so anything outside the two shortlists is a manual review.

2. Map each ask to a question. get_content_opportunities returns the questions where competitors are named and you are not, and those where you appear in under 30% of answers, both capped at 10 rows. Every ask must land on one of those rows or on a shortlist. The budget category is your reading, not a tool field.

3. Read what the answers say. get_prompt_answers lists a shortlisted prompt's recent answers, truncated to 500 characters; get_answer_details opens one in full, with who is named, in what order, citing which sources.

4. Find the duplicates. Flag every prompt or outcome claimed by more than one team, and separately each ask landing on no row.

5. Consolidate. One table: team lead, original ask, consolidated ask, evidence, owner, measurement plan, plus a short message per lead.

What to do with the output

Order the contested prompts by the priority score on each attack and defend row, highest first. One owner per prompt, by this rule rather than by channel: the owner is the team that can measure the prompt once the work ships; if both can, the team whose planned action matches what the answers already cite, an article where the citations are pages you publish, outreach where they are not; if neither fits, the prompt stays out of the cycle.

An ask landing on no row has no evidence behind it. Send it back rather than arbitrating blind.

If both shortlists come back empty there is nothing to arbitrate against: the prompt set is too narrow or too settled, so widen it before the budget conversation.

None of this prices a question. Set the envelope yourself; the output decides what goes inside it and whose name is on each line.

The Complete Prompt

I am consolidating AI-related budget asks from several marketing teams. List the brands I monitor and, if there is more than one, ask which to use before pulling data.

Before step 4, ask me for the asks themselves: team, amount requested, promised outcome. Do not invent them.

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

1. Run prompt signals with limit 10 over the last 30 days. Show the four category counts, aggregate openness, visibility rate, share of voice, and the top attack and defend prompts with their priority scores. Do not produce row-level monitor or ignore lists: only counts exist.
2. Pull content opportunities for the last 30 days, both lists capped at 10 rows. Map each returned question to a required action and a spending bucket: content, editorial, technical, PR or review management.
3. For the shortlisted prompts, list the answers, then open the most relevant in full and report which competitors are named, in what order, and which sources are cited.
4. Cross-reference each ask against those lists. Flag every prompt or outcome claimed by more than one ask, and separately each ask landing on no row.
5. Build one table: team lead, original ask, consolidated ask, evidence, owner, measurement plan. Then a three-sentence message per lead covering the decision, its evidence and the ownership scope.

Apply the ownership rule literally. Order contested prompts by priority score, highest first. The owner is the team that can measure the prompt after the work ships; if both can, the team whose planned action matches what the answers cite; if neither fits, mark it unowned and drop it this cycle. Asks with no evidence row go back unarbitrated.

If step 1 returns zero attack and zero defend prompts, stop there, say the prompt set is not discriminating yet, and list what would have to be added to it first.

Do not assign amounts. Nothing here returns cost or revenue, and pricing a question is my call.