New at ContentCon: meet Contentstack Canoe. See what AI says about your brand.
Contentstack Canoe
Log in Start free
Start free Menu

Acting on recommendations

Up to ten ranked fixes per run, each with its measured reason and one to three steps, plus a playbook for every category.

How recommendations are made

The list is rebuilt every run. The numbers decide what's wrong; a model only writes the steps.

  1. Gaps are detected in code. Each category checks the run's numbers for a clear, measurable gap.

  2. Steps come from the evidence. A model writes one to three from what the engines said and cited for that gap.

  3. Steps are checked before you see them, so they stay within the evidence and off competitors' properties.

Fixes rank by likely impact, and no single category crowds out the rest.

The rules every step follows

  • Only channels you own or can earn: your site and docs, review sites, directories, communities, PR, guest posts and partners.
  • Nothing on a competitor's property, and no invented domains, statistics or claims.
  • Specific to the engine with the gap and to the monitor's persona and region.
  • No shortcuts that don't earn answers, like keyword stuffing or schema tricks.

Whether AI crawlers can reach your site is a separate check: the site audit, on every plan.

Reading a recommendation

2 · CitationsNew Targets citation share

Earn citations from the sources AI trusts

Why?Only 12% of the citations AI gives you point to your own pages. Engines lean on ridereview.example and commutermag.example.

How
  • Get Tarnwell City into the next ridereview.example commuter roundup, with your cold-weather range test results.
View in Citations
  1. Rank and category. Its place in the list and which playbook applies. New: first seen this run.
  2. Target. The metric this fix should move.
  3. Title. The gap in one line, in a fixed pattern per category.
  4. Why? The measured reason.
  5. How. One to three steps from what the engines said and cited.
  6. View in. Opens the evidence.
Anatomy of a recommendation card. Fictional brand and sources.

Overview shows the first three; the full list is under What to do next. Free shows your top two, and the rest are locked until Growth. Directional cards come from thin runs (few engines or questions), so treat them as leads to confirm on the next full run.

Categories at a glance

In typical rank order, though a big gap can outrank a smaller one above it.

Citations

Card titleEarn citations from the sources AI trusts
Shows up whenYour own pages are a small share of the sources cited about you. The smaller the share, the higher it ranks.
TargetsCitation share
Playbook and evidence
EvidenceCitations tab: Citation Share by Owner, Citations by Engine and Sources by Engine
Playbook
  1. In Sources by Engine, list often-cited third-party domains that never point to you. Start with the engine that cites you least.
  2. By type: get listed and reviewed on review sites and directories, pitch publications something new (like test data), answer community questions and build partner pages.
  3. Make your pages citable: the answer in the first paragraph, then specs, prices, test results and a date.
  4. Re-run once placements are live and check New and lost sources.
It worked whenYour share rises and the domains you targeted show up as newly cited.
Tip

The citation targets skill turns this tab into a ranked outreach list.

Competitive

Card titleClose the gap with competitor
Shows up whenA competitor leads you on share of voice, or wins head-to-head questions that don't name you. A bigger lead ranks higher, and close rivals can share one card.
TargetsShare of voice
Playbook and evidence
EvidenceCompetitors tab, plus the head-to-head questions the card quotes
Playbook
  1. On each head-to-head question, read why the engines prefer them. It's usually stated outright.
  2. Publish a fair comparison page: who each option suits, specs side by side, where each wins. Engines lean on advice, not ads.
  3. Earn the proof they're credited with (reviews, case studies, awards) on the sources the engines cite.
  4. Add an alternatives page for buyers who start from their name.
It worked whenYour share of voice rises, you appear on those questions and their line flattens in Brand Ranking Trends.

Funnel

Card titleStrengthen your stage-stage content
Shows up whenOne buyer stage clearly trails your strongest. The wider the gap, the higher it ranks.
TargetsVisibility at that stage
Playbook and evidence
EvidenceVisibility by Funnel Stage, and Prompts filtered to that stage and Gaps
Playbook
  • Awareness. These questions never name you, so answer the category: "best e-bike for a hilly commute" guides, how-to-choose pages and third-party roundups.
  • Consideration. Comparison, alternatives, pricing and migration pages.
  • Decision. Reliability, support, warranty, clear pricing, reviews and case studies.
It worked whenThat stage's ring rises without the others falling.

Prompts

Card titleWin the prompts where you're invisible
Shows up whenNo engine names you on some of your questions. The more there are, the higher it ranks. The card quotes the first ones.
TargetsPrompt coverage
Playbook and evidence
EvidencePrompts tab, STATUS set to Gaps: rows with no checks
Playbook
  1. Copy the exact questions. Each one is a content brief.
  2. Find the page that answers each one plainly. Often there isn't one, or it's buried.
  3. Write or extend it: the answer first, evidence second, in the buyer's words.
  4. Link it from related pages, and get the answer onto the sources the engines cite for that question.
It worked whenThose rows pick up checks on one engine, then more.
Tip

The content briefs skill drafts a brief for each question you're invisible or weak on.

Sentiment

Card titleReframe how engines describe topic
Shows up whenThe engines frame a topic negatively. A verbatim quote behind it ranks it higher.
TargetsBrand sentiment
Playbook and evidence
EvidenceSentiments tab: Topics by Sentiment and the concerns in Sentiment Drivers
Playbook
  1. Trace the quote: expand a question where it appears and check what that answer cites.
  2. Out of date? Post a dated update on the product page and release notes, and ask the cited source to refresh.
  3. True? Answer it with specifics: test data, a fair comparison or a customer story.
  4. Keep publishing. If you go quiet, buyers only hear what other sources say about you.
It worked whenThe label moves toward Neutral or Positive and the concern drops out of Sentiment Drivers.

Strength

Card titleAmplify your lead on topic
Shows up whenThe engines clearly credit you on a topic. It ranks below any gap.
TargetsShare of voice
Playbook
Playbook
  1. Keep the pages that earn the credit current (the ones cited on those questions).
  2. Extend to neighboring questions: win on ride comfort, then answer comfort for longer commutes.
  3. Bring the strength into comparison pages where you trail.

When the list is empty

If the tab says no standing gaps cleared the bar, that's a good result. Keep runs going so new gaps surface fast, and build on Strength cards.

Turn recommendations into work

A list nobody owns doesn't move a number. A routine that works:

  1. Pick one or two per run, from the top unless a lower card is much cheaper to ship.

  2. Give each an owner and a date, with the Why and How pasted into the ticket.

  3. Ship, then re-run with Run now, the on-demand webhook or trigger_geo_test_run. Re-runs use prompts.

  4. Check the deltas. Earned coverage can take a few runs to show.

Let an agent draft the fixes

On Growth and Enterprise, agents read recommendations over the MCP server. Use the Contentstack Canoe skills, or build an Agent OS content agent that drafts entries for review. Nothing publishes without your approval.