What to build — from competitor gap to ranked content plan

This produces a decision, not a keyword dump. Four hundred keywords is not an answer. "Build
these four pages, in this order, in these formats" is.

Workflow

1. Find the gaps

geo_content_gaps(domain, competitors)

or the underlying Labs call, domain_intersection(target, competitor).

Keep every field it returns — each drives a decision below: search_volume, keyword_difficulty,
main_intent, serp_item_types, etv, backlinks_info.

2. Decide the format — never skip this

search_intent maps directly to a content format:

Intent	Build
Informational	Guide, tutorial, explainer, FAQ
Commercial	Comparison, review, alternatives page
Transactional	Product, service, landing page
Navigational	Brand or product page

Then let the SERP elements refine it:

Element present	Action
Many video elements	Make a video, or a video-supported article
people_also_ask	Add a Q&A section
images	Add original visuals with real alt text
featured_snippet	Structure a concise answer, list, or table near the top
ai_overview	Mirror the topics its sources cover
ai_overview and we're retrieved but not cited	Not a volume problem — a directness problem. Rewrite the direct answer.

That last row is the insight worth the whole workflow. A retrieved-but-uncited page is rejected for
being indirect, not for being short.

3. Check we don't already rank for it

Pull current rank tracking first. Never recommend a keyword the user already ranks #1 for — it's
the fastest way to lose trust.

4. Add the AI-demand signal

ai_keyword_volume(keywords, location, language)

Compare AI search volume against Google volume. A topic with high AI demand and low Google demand
is invisible to classic SEO tools and often the cheapest traffic available. Flag these explicitly —
for most sites this is the most valuable output of the whole run.

5. Prioritise

Score: (volume × intent weight) ÷ (difficulty + effort), boosted by rising AI demand.

Return the top 5–10, ordered, one line of justification each.

Never return more than 10. A ranked short list is advice. A long list is work you just handed
back to the user.

6. Write the briefs

For the top 3: target query, format, the questions it must answer, the entities to mention, the
internal links to include.

If an LLM tool is available, generate_sub_topics ($0.0001/task) turns a topic into an outline
cheaply. Prefer it to a full draft — clients usually want to write it themselves.

Reporting

One page. A ranked table: query · format · why now · effort. Then briefs for the top 3.

State the cost of the underlying data calls before a large sweep.

Never

• Never recommend "publish more content" as a specific action.
• Never recommend a keyword the user already ranks #1 for.
• Never present AI demand as search volume — different metrics, different meanings.
• Never give an unranked list of more than 10 items.