Audits

An audit puts a brand's selected prompts to the selected AI platforms, scores every answer, checks the answers against the brand's facts, and turns the result into recommendations. This page is the reference for what goes in, how a run behaves, and what comes out.

Inputs

Input	Rule
Brand	The active brand. The New Audit form has no brand picker; switch brands in the header first.
Prompts	Active prompts only. Every active prompt starts selected on the form; "Skip N audited in the last 10 days" drops the recent ones. Each prompt is 1 to 300 characters on the server; the dashboard's add box stops at 100.
Platforms	Any subset of the plan's platforms. At least one prompt and one platform must be selected.
Prompt text	Each prompt is sent exactly as written. The exact text is stored with each answer and shown under "Prompt sent".
Brand facts	Read at run time for the accuracy check. None on file means accuracy is reported as not checked.
Competitors	Tracked competitors are looked for in the same answers.

How each platform is asked

Key	Platform	Retrieval in our audit	How it is asked	Plans
`chatgpt`	ChatGPT	On	OpenAI API, model `gpt-6-luna`, with web search on. The model decides per prompt whether to search, as ChatGPT does, and each answer records whether it searched.	Free, Starter, Growth, Pro
`aio`	Google AI Overviews	On	The AI Overview Google shows for the prompt on a US, English-language, desktop search, read from the results page through DataForSEO. When Google shows no AI Overview, the prompt is skipped for this platform, not scored as a miss.	Free, Starter, Growth, Pro
`ai_mode`	Google AI Mode	On	The answer Google AI Mode gives for the prompt on a US, English-language, desktop search, read through DataForSEO, with the sites it drew on. AI Mode answers every query, so a missing answer is reported as not measured.	Starter, Growth, Pro
`google`	Gemini	On (search grounding)	Gemini API, model `gemini-3.8-flash`, with Google Search grounding. If the grounded call fails it is retried once without grounding, so an occasional Gemini answer is ungrounded.	Starter, Growth, Pro
`perplexity`	Perplexity	On	Perplexity API, model `perplexity/sonar`, with web search.	Starter, Growth, Pro
`claude`	Claude	On	Anthropic API, model `claude-haiku-5-5`, with web search on (up to three searches). Claude decides per prompt whether to search, and each answer records whether it did.	Starter, Growth, Pro

Retrieval matters when you read results. An answer from a platform with retrieval off reflects what the model learned in training; it cannot see a change you made last week.

Lifecycle

Status	Meaning
`running`	Created and claimed against the monthly allowance. Execution starts when the run request reaches the server.
`pending`	Only after a staff reset. Never aged out automatically.
`completed`	Scored and stored.
`failed`	Stopped without a score. The reason is stored but not shown in the dashboard.

Time bound. One run executes inside a single 300-second server function. Platform calls that have not returned by the internal deadline are dropped and the audit is scored on what came back.
Partial results. A platform that errors or skips a prompt leaves that answer unmeasured, with a reason. Only measured answers enter the score.
All platforms failed. The audit is marked failed.
Stuck runs. A started audit still running after 10 minutes is set to failed with "Audit timed out and was marked failed automatically." It happens on the next read of the audit, or in a nightly cleanup.
Allowance. Only completed and running audits count toward the month, so a failed audit returns its slot.
Spend limit. If the platform-wide daily or monthly AI spend limit is reached, the audit fails with SPEND_CAP_REACHED (HTTP 503) and the slot is returned.

What is measured per answer

mentioned and mention_count: whether, and how often, the answer names the brand (name variants, possessives and the domain).
sentiment and sentiment_score: tone, from counts of positive and negative words, weighted toward words near the mention.
citations: each link with url, title and is_own_site. Owned URLs on the brand count as own site.
Accuracy flags against the brand facts. See Hallucination detection.

Outputs

Overall score, per-platform scores and per-prompt scores. See Visibility score.
accuracy_checked, false when the brand had no facts.
Up to 15 recommendations. See Recommendations engine.
Competitor mention rates, share of voice and prompt gaps, when competitors are tracked.
Citation sources by domain, and citation gaps on Starter ($49/mo) and above.
Google positions for the same prompts on Starter ($49/mo) and above; the backlink profile on Growth ($99/mo) and above.

Limits

Free — 1 audit a month
Starter ($49/mo) — 15 audits a month
Growth ($99/mo) — 50 audits a month
Pro ($249/mo) — 150 audits a month

The allowance belongs to the account, not the brand: every brand draws on one pool. It resets on the 1st of the month. Unused audits do not carry over.

Scheduled re-audits

Each brand has a cadence: off, weekly, monthly. Scheduling is on Starter and above, set under Brands → Edit → Re-audit schedule.

A scheduled run asks about every active prompt on every platform the plan allows.
Runs are dispatched once a day, in the morning UTC. The first comes one full interval after the schedule is turned on.
Each run uses one audit from the allowance and is skipped when none are left. The schedule switches off if the plan no longer allows it.
Each run sends a re-audit email that leads with what changed.

Notifications

Audit complete: subject "Audit Complete: [brand] scored [score]/100", sent when an audit you start finishes. On by default; turn off under Settings → Email Notifications.
Accuracy alerts: sent when an audit, started or scheduled, finds a critical flag.
No email is sent when an audit fails.

Errors an audit can return are listed in Error messages. For a stuck or failed audit, see Audit stuck on Running or failed; to get a first audit started, Can't run your first audit.

Quick Start

Visibility Score