Skip to main content
LLM Metrix
AI · Recommendations

AI-built action plans.
Not just dashboards.

LLM Metrix doesn't just tell you what's broken. It turns each scan into GEO recommendations, content edits, schema additions and citation targets, prioritised critical through low, so generative engine optimization becomes a backlog you work, not a report you file.

Start freeBook a demo

Free forever plan. No card required

app.llmmetrix.com/dashboard/recommendations
Illustrative, sample figures in the product's real layout
Action plan
4 AI-generated tasks
Completed
0/4

Click a task to mark it complete as you work through your action plan.

Works with the AI engines your customers use

ChatGPTPerplexityGeminiClaudeGrokMeta AIDeepSeekGoogle AI OverviewsMicrosoft Copilot

Why it matters

From finding to shipped fix.

An action, not an adjective

No vague advice. Each item names the change to make and the scan finding behind it, and the entity ones carry the exact sameAs JSON-LD to paste.

Prioritised for you

Each task carries a priority, critical through low, and the board sorts by it, so the team always works the highest-priority items first.

A brief in one click

Every recommendation is a brief you can hand over: the change to make, why the scan called for it, and which engines it targets.

Schema guidance

Recommendations flag where structured data like FAQPage or Product schema could help you win the prompts you care about.

Track your progress

As new scans come in after you ship changes, watch your visibility score respond so you can gauge what's working.

Continuously refreshed

Recommendations refresh after every scan as your score, your weakest engines, the prompts you were missed on and your competitor set change.

Mechanics

From completed scan to shipped fix.

  1. 01

    Track the questions worth answering

    Add the prompts buyers actually type, or generate candidates with discovery and pick from topic clusters. Tracked prompts stand permanently and re-run on every scan rather than being consumed once.

  2. 02

    Derive the gaps into items

    After each completed scan, a precompute turns the results into board items, engines you are weak on, prompts you were missed on, competitor divergence, community sources, each tagged critical through low.

  3. 03

    Reconcile the board, never duplicate it

    Items carry a stable key, so a gap that persists keeps its status across rescans while one that clears auto-resolves, and reopens if it comes back. Done and dismissed rows stay as history.

  4. 04

    Generate the artifact the row warrants

    An eligible row can produce an llms.txt file, schema JSON-LD or a content brief, copied or downloaded straight from the card; generating one moves the row to todo and writes an audit-trail entry.

  5. 05

    Measure shipped work on the next scan

    Done items keep a frozen evidence snapshot that the next scan's fresh facts are measured against, while your 0–100 visibility score plots per completed brand scan, the loop closes on measurement.

Who it's for

Who turns findings into fixes

Content lead working a prioritised backlogA sprint board for visibility, instead of another PDF audit.

You own the content roadmap and need next week's work decided by something firmer than whoever argued hardest.

  • Items are tagged critical through low, and the board orders the queue by that tag.
  • Every card states the change to make, the scan finding behind it and the engines targeted, a brief, not a hint.
  • Content-gap rows generate a content brief you can hand straight to a writer.
  1. 01Run a scan so the board fills itself from that scan's gaps.
  2. 02Work the critical-tier rows first, the queue is ordered for you.
  3. 03Generate a content brief from a content-gap row and hand it to a writer.

Metrics this role tracks: Priority tiers · Board items done · Own-domain citations

Open your backlog
SEO engineer shipping schema without guessingPaste-ready markup where the scan found the hole.

You implement the technical fixes and want the spec written out, not described at you in prose.

  • Technical rows unlock generated schema JSON-LD, script tag and validation notices included.
  • Entity recommendations ship the literal sameAs JSON-LD block with real URLs interpolated.
  • Eligibility is mechanical: the row's category decides which artifacts are offered, on the card and at the route alike.
  1. 01Open the Recommendations board and take the technical-workstream rows.
  2. 02Generate the schema JSON-LD from an eligible row and copy the script tag.
  3. 03Move the row to done once deployed so the next scan measures the outcome.

Metrics this role tracks: Board items done · Priority tiers · Citation volume trend

Generate schema JSON-LD
Agency deliverable owner showing client progressProgress the client can see without a status meeting.

You report monthly across retainers and need shipped-work receipts rather than activity summaries.

  • Board status is shared workspace state, whoever ships an item moves it, and every teammate sees the same board.
  • Done rows keep their frozen evidence snapshot, measured against the next scan so the outcome lands on the row.
  • Resolved, done and dismissed items persist as history, so the record of what changed never evaporates between reports.
  1. 01Run a scan per client project so each board stays scoped to that brand.
  2. 02Mark rows done as your team ships them to build the outcome record.
  3. 03Walk the client through the done rows' measured outcomes at report time.

Metrics this role tracks: Board items done · Priority tiers · Own-domain citations

Show the client board
Solo marketer with no GEO specialistThe whole GEO function, minus the hire.

You are one person covering SEO, content and product marketing, and AI visibility landed on your desk anyway.

  • Briefs explain why, not just what, the rationale text is readable without specialist vocabulary.
  • Visibility rows offer an llms.txt file generated for your domain, validated before you download it.
  • Weekly automatic scans on free keep the board current while you work it in priority order.
  1. 01Run your first scan, recommendations appear without any setup.
  2. 02Generate the llms.txt file offered by a visibility row and upload it to your site root.
  3. 03Snooze anything you cannot do this month so the top of the board stays honest.

Metrics this role tracks: Priority tiers · Board items done · Citation volume trend

See what engines missed
Category manager owning a product line's entity presenceYour line's facts kept straight wherever engines answer for the category.

You own a product line end to end and engines keep describing it loosely, wrong category, missing specs, a rival's framing.

  • Entity recommendations carry the literal sameAs JSON-LD, interpolating your real URLs so the product line links to the right profiles.
  • Schema rows unlock generated JSON-LD copied or downloaded straight from the board card.
  • Workstream tags route each fix, technical, content or PR, so your line's items land with the right team.
  • Each item states which engines it targets, so roadmap arguments stop at the card.
  1. 01Run a scan covering the prompts buyers ask about your product line.
  2. 02Take the entity rows carrying sameAs JSON-LD to whoever ships site changes.
  3. 03Mark each implemented item done and let the next scan measure the outcome.

Metrics this role tracks: Priority tiers · Board items done · Own-domain citations

Open your product's board

Foundations

What is generative engine optimization (GEO)?

Generative engine optimization is the practice of improving how AI systems understand, represent and recommend your brand. The term comes from a 2023 research paper that defined generative engines as systems answering a query by synthesizing across multiple retrieved sources, and it has since become the umbrella over two related disciplines: AEO, focused on getting quoted inside the answer, and the broader work of shaping what the model itself knows about you.

Engines know your brand through two channels. Training data is absorbed slowly but durably; live retrieval at question time moves in days. Both respond to the lever that paper measured, content that added quotations, statistics and cited sources gained visibility, while keyword-styled edits did not. That finding shapes everything this feature measures: your tracked prompts are the questions buyers actually ask, and each scan shows which side of the answer your brand lands on.

Practice

How do you get cited by AI engines?

Retrieval-first engines cite what they can find, and they quote what is easy to lift. In practice that means pages stating facts plainly enough to attribute, structured data such as FAQPage or Product schema that lets an engine quote you precisely, and an llms.txt file that points crawlers at the pages that matter. None of it is trickery, it is making the correct answer easy to borrow.

The second half is placement. Engines lean on sources they already trust, so presence in the third-party domains cited for your prompts, trade press, documentation hubs, community threads, often opens the door faster than another page on your own site. Your recommendation board is generated from exactly these signals: which engines missed you, which prompts went unanswered, whose sources crowded you out.

  • State facts attributably, quotable sentences survive retrieval better than adjectives.
  • Structured data helps engines quote you precisely rather than paraphrase loosely.
  • Presence in already-cited third-party domains compounds faster than isolated publishing.

The board

What happens to a recommendation after each scan?

Recommendations are keyed stably, so the same gap re-derived from a later scan keeps its board status instead of arriving twice. Open items whose underlying fact cleared in the latest scan resolve automatically; if the gap returns, the item reopens rather than staying quietly fixed. Snoozed items come back when their window lapses, and done and dismissed rows remain as history.

Done is not the end of the record either. A completed item keeps a frozen snapshot of the evidence that justified it, and the board measures that snapshot against freshly derived facts on the next scan, so the before-and-after of shipped work is read off real results instead of remembered from a meeting.

FAQ

Recommendation questions, answered.

How are recommendations generated?
We feed your project's own scan results, your visibility score, the engines you are weakest on, the prompts you were missed on and the competitors named instead, to an AI model that returns a list of recommended edits, each with a priority. Everything is generated from your own monitoring results.
How are recommendations prioritised?
Each recommendation carries one of four priorities, critical, high, medium or low, assigned by the model that wrote it, from the gap in your scan it addresses. Priorities are guidance to help you sequence the work, not a predicted score change.
What kinds of changes do GEO recommendations cover?
Three families: content work (pages to write or expand around prompts you were missed on), structured data (where FAQPage, Product or similar schema could help engines quote you), and entity and citation work (connecting your brand to the right facts and sources). Each item states the change to make, why the scan called for it, which engines it targets, and a priority from critical through low.
When do recommendations update?
After every scan. A completed scan triggers the recommendations precompute, so your board reflects the latest results, refreshed daily on paid plans and weekly on free from automatic scans, plus any scan you run yourself.
Do you rewrite our pages for us?
No, you get the brief, not replacement copy. Each recommendation names the change to make, why the scan called for it and which engines it targets, which is what a writer or engineer needs to execute it well. The one paste-ready exception is entity work: those recommendations include the exact sameAs JSON-LD snippet to add to your site.
How do recommendations relate to my visibility score?
The score is the measurement; recommendations are the work. Your visibility score is plotted for every completed scan of your brand, and each scan's gaps generate the items on your board. Ship the fixes, let later scans run, and the same trend shows whether things moved, measurement and action in one loop.
Can a recommendation become a ready-to-use file?
Yes. An eligible item on your recommendations board can be turned into a generated artifact, an llms.txt file, schema JSON-LD or a content brief, copied or downloaded straight from the card. Every generation is recorded in the audit trail, so you can see what was produced and when.
How do we get started with GEO recommendations?
Run your first scan, recommendations are generated from its gaps automatically, so the board has work on it before you write anything. Work items in priority order, move them along as you go and mark them done when shipped; later scans re-check the facts behind open items against fresh results.
Can I sync recommendations to Linear or Jira?
Integrations with tools like Linear, Jira, and Notion are on our roadmap. Today you can work the prioritised recommendations directly in your LLM Metrix dashboard.

Keep exploring

Related features

Visibility Score

A single composited score from mention rate, answer position and sentiment that shows how AI engines understand and recommend your brand.

Citation Intelligence

Discover which domains AI engines cite in answers about your space, and the exact page when the engine returns a link.

Brand Safety Monitoring

Flag potentially inaccurate descriptions and harmful brand associations on each completed scan, for review.

In depth

The Citation Authority Playbook: How to Get AI to Cite Your Brand

Being cited by AI engines isn't luck. It's the result of a deliberate content strategy. This playbook covers structured data, authority signals, and the distribution tactics that actually work.

13 min read

Does llms.txt Actually Do Anything? An Evidence Check

A proposed standard with real adoption, genuine appeal, and, as of today, no major AI engine publicly confirming it consumes the file. Here's what it is, what the evidence supports, and a verdict on whether to publish one.

7 min read

Does Schema Markup Help AI Visibility? Separating the Two Claims

Structured data is either essential or irrelevant to AI answers depending on who you ask. Both camps are answering different questions: one about Google's surfaces, one about what a language model reads. Here's the split.

8 min read
Start on the free plan. No card, no time limit.

The AI era of search
is already here

See what AI engines say about your brand before your competitors do. Start free today. No card required.

Start freeBook a demo