Skip to main content
OnPageIQ Documentation

AI Visibility (LLM Listening)

Updated · Aug 19, 2026

See how the answer engines describe your brand — overall, per engine, and per place you serve — with the evidence behind every number.

What changed

Local AEO Visibility was retired into this add-on rather than shipped as a second product. New here: a per-project local mode with its own Locations tab; the Overview and Citations split by place; opportunities that feed the existing Intelligence recommendations instead of a second inbox; AI-crawler activity read from your own Cloudflare zone (a leading indicator, kept apart from visibility); and a shareable report, on screen and as a co-branded PDF.

User goal

Compare how ChatGPT, Claude, Gemini, and others answer prompts about your brand.

Tier
Enterprise default or Stripe add-on
Nav label
AI Visibility
Route
/projects/{project}/llm-listening/keyword-sets
Gates
llm-listening-addon llm-listening-addon

Prerequisites

  • llm-listening-addon active
  • Provider keys or org BYOK configured
  • Credits for run execution

QA focus

  • Runs scoped to project keyword sets
  • Progress tracker updates via broadcast
  • Results master-detail shows all providers
  • Citations lists rows after a run that cited sources — an empty Citations page alongside a populated Overview means the roll-up did not record them

Overview

AI Visibility tracks how ChatGPT, Claude, Gemini, Grok, and other models answer prompts about your brand — side by side, with extracted mentions and run history.

What you get

  • Keyword sets for prompts, brand terms, and run history
  • Parallel runs across the models you select, with a live cost estimate before launch
  • Real-time progress while queries finish
  • Master-detail results to compare answers and brand mentions
  • Citations — every source the models cited, grouped by domain and classified as your own site, an authority reference, a community forum, or other third-party coverage
  • Claims — the factual statements models make about you, including ones they cite no source for
  • Optional Suggest with AI from the project Business Profile
  • Provider keys and usage under AI Visibility settings

If you serve particular places

Turn on local mode and the same questions get measured from each place you list:

  • A Locations tab for the cities you serve, with a cap on how many are measured per run
  • The Overview and Citations split by place, so "we are invisible in Boston" becomes a number rather than a hunch
  • Locations that no search can be aimed at are called out rather than shown as zero visibility

What to do about it

  • What to fix — questions the engines answer without naming you, questions where you are named but never the pick, and places measurably behind your own average. These land in Intelligence alongside your search and analytics recommendations, not in a second list
  • Bot activity — how often GPTBot, ClaudeBot and the rest fetch your pages, read from your own Cloudflare account. A leading indicator: it shows the engines can read you, never that they mention you
  • Report — a shareable summary of all of the above, on screen and as a PDF carrying your agency's branding

How it works

  • Create a keyword set, add prompts (or accept AI suggestions), then launch a run
  • Snapshots store each keyword × model answer and extracted mentions
  • When a run finishes, its answers are rolled up: the sources they cited become the Citations list, so that page reflects the most recent run that actually cited anything
  • Export results when you need an offline copy; budget and daily run limits can block a launch with a clear message
  • Every figure is reported as a count with its confidence ("mentioned in 3 of 12 answers"), never a bare score. Answers that errored or were refused are excluded from every denominator

Common issues & false alarms

  • Horizon must be running to execute the queries; live progress streams over Reverb and falls back to a periodic refresh if Reverb is unavailable (no manual reload needed)
  • Budget exceeded → run blocked with message
  • Citations empty after a run that clearly cited sources → the run roll-up has not recorded them; runs that finished before the fix do not backfill themselves, so re-roll them with `php artisan llm-listening:backfill-citations`
  • Brand terms are capped at 100 per set, 100 characters each — pasting a longer list adds only what fits
  • Local mode on with no locations added → runs fall back to a single national measurement, and the Overview says so rather than showing an empty chart
  • A location with no city and no coordinates cannot be aimed at, so runs record a skip for it — the Locations tab flags it as "Cannot be measured" rather than reporting zero
  • Only the Google AI Overview capture carries a location; the chat engines answer the raw question, so the location breakdown reads from the AI Overview roll-up
  • A Cloudflare token that has expired shows as "Failing" with the reason, never as a week with no crawler visits

Interactive guide

Step of

All steps (reference)

  1. Step 1. Create keyword set

    Container for prompts, brand terms, and run history.

    What to do: llm-listening.keyword-sets.create → save set

    Where: llm-listening.keyword-sets.create

    Create keyword set

    Expected (pass)

    • Set appears on index
    • Project binding enforced on save

    Negative cases (must fail safely)

    • Flag off → llm-listening.upgrade or 403

  2. Step 2. Suggest keywords with AI (from the Business Profile)

    On a keyword set, open "Suggest with AI" to turn the project's Business Profile into suggested SEO keywords, LLM prompts, and brand terms. Review, edit a term inline before accepting, select all high-confidence, or dismiss ones that do not fit (with Undo); accepting adds them to the set respecting the lock and tier cap. Requires an AI Business Profile — build it at Settings → Business Profile.

    What to do: Open a set → Suggest with AI → select/edit → Add selected

    Where: llm-listening.keyword-sets.index

    Suggest keywords with AI (from the Business Profile)

    Expected (pass)

    • Accepted terms added respecting set lock + tier cap
    • Duplicate/over-cap items reported honestly

    Negative cases (must fail safely)

    • No ready profile → CTA to generate one

  3. Step 3. Add keywords and launch run

    Bulk editor for prompts; launcher selects models and starts run.

    What to do: keywords.bulk-editor → add prompts → launch run

    Where: llm-listening.keywords.bulk-editor

    Add keywords and launch run

    Expected (pass)

    • Run status transitions queued → running → completed
    • Cost estimate shown before launch

    Negative cases (must fail safely)

    • Over daily run limit → validation error

  4. Step 4. Track progress

    Real-time run progress with per-keyword/model status.

    What to do: Open llm-listening.runs.progress during active run

    Where: llm-listening.runs.progress

    Track progress

    Expected (pass)

    • Progress UI updates without full page reload
    • Failed snapshots show error state

  5. Step 5. Compare results

    Master-detail view of answers side-by-side with brand mention extraction.

    What to do: llm-listening.results.show with runId

    Where: llm-listening.results.show

    Compare results

    Expected (pass)

    • All selected models have snapshot rows
    • Export modal produces file when used

    Negative cases (must fail safely)

    • Foreign keywordSet → 404

  6. Step 6. See which sources the models cited

    Citations lists every source behind the run, grouped by domain and classified as owned (your own site), authority, community, or earned — so you can see whether models are reading you or reading about you somewhere else. Rows are populated by the run roll-up, so the page shows the most recent run that actually cited sources, and says so when that is not the latest run.

    What to do: AI Visibility → Citations → filter by source type

    Where: llm-listening.citations.index

    See which sources the models cited

    Expected (pass)

    • Each cited domain appears once, with its citation count and a sample URL
    • Source-type filters narrow the list
    • A run with no cited sources leaves the page empty rather than erroring

    Negative cases (must fail safely)

    • No run has cited a source yet → empty state, not a crash

  7. Step 7. Measure each place you serve

    Turn on local mode for the project, then list the cities it serves. Every future run measures each one separately instead of producing a single national average that describes none of them.

    What to do: Project Settings → Local SEO → set the mode to Local, then open AI Visibility → Locations and add a city

    Where: llm-listening.locations.index

    Measure each place you serve

    Expected (pass)

    • Locations tab appears only for a project in local mode
    • A location with no city and no coordinates is flagged "Cannot be measured"
    • Retiring a location keeps everything it has already measured

    Negative cases (must fail safely)

    • Broad-mode project → no Locations tab and no change to any existing screen
    • At the per-project cap → adding is refused, editing an existing location still works

  8. Step 8. Read the per-place breakdown

    The Overview gains "How each location sees you". Every active location appears, measured or not — a place that produced no answer is shown with the reason rather than left out, because a missing row reads as "no visibility here" when the truth may be "this was never asked".

    What to do: Open AI Visibility → Overview on a local project after a run completes

    Where: llm-listening.overview

    Read the per-place breakdown

    Expected (pass)

    • Each location shows its own rate with its own confidence
    • Unmeasured locations state which of the three reasons applies

    Negative cases (must fail safely)

    • Broad-mode project → panel absent entirely

  9. Step 9. Filter citations by place

    Citations lists every source the engines cited — now including Google AI Overview captures — and a local project can narrow it to one place.

    What to do: Open AI Visibility → Citations and pick a location chip

    Where: llm-listening.citations.index

    Filter citations by place

    Expected (pass)

    • A chip appears only for a location this run actually cited
    • Retired locations stay in the filter so the counts still add up

    Negative cases (must fail safely)

    • A location id from another organization in the URL → ignored, list stays unfiltered

  10. Step 10. See which AI crawlers read the site

    Connect a read-only Analytics token from your own Cloudflare account and Bot activity counts how often the AI crawlers fetch your pages. Aggregate counts only — no IP addresses are requested or stored.

    What to do: Project Settings → Integrations → Cloudflare → paste an API token and zone ID, then open AI Visibility → Bot activity

    Where: llm-listening.bots.index

    See which AI crawlers read the site

    Expected (pass)

    • Saving checks the credentials immediately and reports the outcome
    • The page states that crawler visits are a leading indicator, not visibility
    • A failing connection shows the reason, never an empty week

    Negative cases (must fail safely)

    • Token rejected by Cloudflare → "Failing" with the fix, and no data claimed
    • Disconnecting keeps the crawler history already collected

  11. Step 11. Share the report

    A summary of everything above, on screen and as a PDF with your agency branding. The on-screen version and the PDF are the same document.

    What to do: Open AI Visibility → Report, then Download PDF

    Where: llm-listening.reports.show

    Share the report

    Expected (pass)

    • Every measure shown as a count with its confidence
    • Crawler numbers always carry the leading-indicator note
    • No rank claims and no traffic attribution anywhere in the document

    Negative cases (must fail safely)

    • No completed runs → the report says so plainly instead of rendering empty tables

  12. Step 12. Check the country you are measured in

    Every measurement is taken in one country's results, and a UK business asking a US search engine gets US answers, US competitors and US citations. You do not have to set this: the country is read from your Business Profile, from a confirmed address, from your Local SEO settings, or from your domain ending — whichever is the most reliable thing we already know about you. This screen is where to correct it if it is wrong.

    What to do: Project Settings → Business Profile → Market → Country

    Where: projects.settings.business-profile

    Check the country you are measured in

    Expected (pass)

    • The country you type here wins over anything the profile inferred
    • A spelled-out name works as well as a two-letter code — "United Kingdom" and "GB" both resolve
    • A business with nothing set keeps being measured exactly as before

    Negative cases (must fail safely)

    • A country we cannot target yet → measurement falls back to the default market rather than failing the run, and the mismatch is logged
    • Changing the country starts a fresh comparison instead of reporting the difference between two countries as a change in your visibility

Related guides

Billing & Subscriptions

Manage subscription tier, purchase credits, add-ons, and view billing history.

Integrations (GSC & GA4)

Connect Google Search Console and Google Analytics 4 per project.

Intelligence Hub

Cross-source insights combining GSC, GA4, scans, and audits.

AI Business Profile & Keyword Assist

Let AI infer your market, audience, competitors, and priority keywords from your crawl — then reuse it in Keyword Research and AI Visibility (LLM Listening).

Local AEO Visibility (retired — see AI Visibility)

Retired. Everything it measured now lives inside AI Visibility, behind a per-project local mode.