Prompting Guide

aeo_research runs Brave-powered web research and synthesises entities, source-backed claims, competitor angles, and data points to ground every downstream tool.

• depth: standard (10cr) for quick factual checks or narrow topics. deep (15cr) for comprehensive multi-angle research with follow-up coverage rounds on broad topics.
• region: Set to a two-letter country code (e.g. us, gb) to localise search results. Omit for global coverage.
• recency: Prefer week or month for trending topics; year for evergreen content.
• brief: A focused brief improves entity coverage. Include the target audience and specific angles you want explored.

aeo_outline converts research data into a structured, AEO-aware outline with headings and section intents matched to your content type.

• contentType (outline / hosted draft): article (long-form narrative), faq (question-first), product (features/benefits), or guide (step-by-step how-to).
• Programme article formats (content menu / room picks — separate from outline contentType): guide, comparison, faq, case_study, announcement, original_research. Skeleton checks expect format-specific evidence (e.g. comparison → table, guide → numbered steps).
• Writing a good brief: Include your target keyword, audience, desired length, and any angles or entities that must be covered. Example: "2000-word guide on container security for DevOps engineers. Must cover runtime scanning, image signing, and SBOM generation."

aeo_write drafts an article section by section from your outline and research, citing only URLs that appear in the research results.

• tone: Provide a short description like "professional but approachable" or "technical and concise".
• readingLevel: Controls vocabulary and sentence complexity. Use general for broad audiences, expert for technical readers.
• Citations: Only URLs found in the research results are used as citations. Fabricated sources are filtered out automatically.
• Grounding gate: aeo_write returns 422 research_required without research-backed sources. Submit-for-review and final saves require inline citations from the research allow-list. Lightweight agents: use aeo_create for a grounded end-to-end draft.

aeo_polish applies a brand-voice editing pass to a draft, refining sentence-level style while leaving headings and structure intact.

• Idempotent: Safe to run multiple times — the tool detects already-polished content and makes minimal changes on subsequent passes.
• brandVoice: Provide explicit style rules. Examples: "Never use passive voice", "Always use Oxford commas", "Capitalise product names but not features".
• Headings and structure are preserved — polish only affects sentence-level style, not document architecture.

aeo_factcheck deterministically extracts statistical claims and checks each against the research allow-list — no LLM, reproducible. Prefer this for numbers and rates.

• Pass research JSON or project_id so claims are checked against the same sources used to write.
• For qualitative / entity claims, aeo_verify (2cr) still provides an LLM-judged review pass.

aeo_score grades content across the 8 AEO dimensions on a 0–100 scale with per-dimension feedback, and it's free to run on every edit.

• Instant scoring across all 8 AEO dimensions plus an overall weighted score with extractability bonus.
• When to score: After every edit. It's free — use it as a feedback loop to see which changes actually improve your content.
• Pass research: Include research (or project_id) so scoring can surface grounding tips and next actions when citations are missing.
• Each dimension gets a 0–100 score. Focus on your lowest dimensions first. For rewrites toward a target, prefer aeo_improve mode to-score.

aeo_audit fetches a URL or markdown, scores it, and returns a prioritised improvement plan with missing entities and schema opportunities.

• With targetKeyword: Adds competitive research context — compares your content against top-ranking results for that keyword.
• Without targetKeyword: Standalone audit that evaluates content on its own merits.
• Accepts a URL (fetches the page) or raw markdown. URLs get richer analysis since the tool can inspect meta tags, schema markup, and page structure.
• Reading results: Focus on improvements (prioritised fixes), missingEntities (gaps in coverage), and schemaOpportunities (structured data you should add).

aeo_compare runs a head-to-head AEO comparison of two pages or drafts and reports the winner and the point spread between them.

• URL vs markdown: URLs get JSON-LD detection and full page analysis; markdown skips schema detection. Prefer URLs when available.
• targetKeyword: When provided, the comparison is contextualised for that keyword — useful for competitive gap analysis.
• marginOfVictory: The point spread between the two pieces. A margin <5 means the pieces are near-equal; >15 indicates clear winner.

aeo_extract reformats content for a specific answer-engine surface — featured snippet, knowledge panel, or People Also Ask.

• Format-specific: Use featured_snippet, knowledge_panel, or people_also_ask to target a specific extraction format. Use all to get every format at once.
• scoreBreakdown: Shows how well your content fits each extraction format. Use this to decide which format to optimise for.

aeo_crawlability checks whether 11 major AI bots can reach a site — robots.txt, meta directives, llms.txt, sitemap, and JS rendering — and returns prioritised fixes.

• Pass any website URL. Returns pass/fail for 11 AI bots, llms.txt presence, sitemap check, JS rendering assessment, and prioritised fix recommendations.
• Best practice: Use as the first step before any audit to catch hard blockers that would prevent AI engines from accessing the content at all.

aeo_llms_txt generates spec-compliant llms.txt and llms-full.txt files for a site, grouping pages into semantic sections with deployment instructions.

• Auto-discovers pages and groups them into semantic sections. Produces both standard llms.txt and extended llms-full.txt.
• Deliverables: Include the deployment instructions in client deliverables — the output contains exact file contents and hosting guidance.

aeo_refresh runs fresh research against existing content and returns a targeted update plan, or reports that the content is still fresh so you don't spend credits needlessly.

• Accepts a URL (fetches latest version) or raw markdown. URLs are preferred for live-content freshness checks.
• content_is_fresh status: If the tool returns this status, your content is already up-to-date — no refresh is needed and no credits are wasted on unnecessary rewrites.
• confidenceLevel: High confidence means the refresh suggestions are well-supported by fresh sources. Low confidence means some suggestions are speculative — review before applying.

aeo_share_of_voice estimates AI answer visibility against competitors via a Brave-search ranking proxy (labelled — not confirmed ChatGPT / Perplexity / Gemini citations) for up to 20 keywords.

• Track up to 20 keywords with the Brave visibility proxy. Compare client domain vs up to 5 competitors. For measured engine citations, use the visibility scoreboard / citation board.
• Returns: SOV %, coverage, gap keywords (competitors appear but client doesn't), and domination keywords.
• Best practice: Run monthly for trend tracking — SOV shifts over time reveal whether your content strategy is gaining or losing ground.

aeo_cluster groups keywords into a semantic content hierarchy and calendar, flagging any keywords the model invented to fill gaps.

• Optimal keyword count: 20–100 keywords is the sweet spot. Fewer than 20 produces trivial clusters; more than 100 increases noise. Hard cap is 200.
• domain: Providing your domain gives the tool context about your site's existing authority and content, improving cluster relevance.
• hallucinatedKeywords: These keywords were invented by the LLM to fill gaps — they were not in your input. Review them: some are useful suggestions, others should be discarded.
• maxClusters: Leave unset for automatic grouping, or set to constrain the output. Useful when you have a fixed content calendar with N slots.

Prefer aeo_create for hosted drafts. aeo_full_pipeline remains as an alias (draft 12cr / publish 20cr).

• Single / gate (20cr): Research → outline → write → polish → grounding. Gate mode iterates toward a target score.
• Strategy (~44cr): Cluster → research + outline top pillars (same economics as aeo_workflow_strategy).
• Pass pre-computed research: If you already ran aeo_research, pass the result so the pipeline can skip a duplicate research spend.