Product overview

One loop: Track, Explain, Improve, Prove.

AnswerRank is built around four questions asked in order. A number without a cause is not actionable, and a change without a re-test is not evidence. Every surface in the product serves one of the four.

Run a free visibility check See how we measure

Unit of measurement: One check = one prompt × one engine × one market.
Engines covered: OpenAI, Claude, Gemini and Perplexity web-grounded APIs, plus observed Google AI Overview SERP data.
Reporting rule: Every metric ships with its sample size, its confidence level, and how it was collected.

The loop

Four questions, in order.

Most visibility tools answer the first one and stop there. The remaining three are where the work actually gets decided — and where a tool either shows its evidence or asks you to trust a number.

01
Track
Does AI mention us?
A check is one prompt × engine × market. AnswerRank runs your prompt library on a schedule and records how your brand appeared in each answer — recommended, listed, mentioned in passing, or absent.
Overview · Prompts · Engines
02
Explain
Who wins instead, and why?
Every gap is traced back to the answers that produced it: which competitors took the slot, which sources the answer cited, and what your own pages did or did not give the engine to work with.
Competitors · Citations · Sources · Site audit
03
Improve
What exactly do we do?
Findings become scoped recommendations with impact, effort and confidence attached, each linked to the prompts and citations that justify it. Nothing is published on your behalf.
Recommendations · Content briefs
04
Prove
Did it work?
Freeze the prompt set, ship the change, re-run the identical checks, and read the delta against a stated evaluation window — with an explicit refusal to call a result on a thin sample.
Experiments · Reports

Track

Answers, prompt by prompt — not a single aggregate.

AnswerRank builds a prompt library from your site and category, classifies each prompt by buying intent, and runs it across every configured engine and market. The row-level record is the product; the score is a summary of it.

Demo dataset

Check log · Acme · United States / English

Illustrative check log for the fictional brand Acme
Prompt	Intent	Engine	Result	Position
best analytics platform for a mid-market SaaS team	Comparison	Perplexity	Recommended	2 of 6
Northwind vs Acme pricing	Transactional	OpenAI	Listed	4 of 5
alternatives to Initech for product analytics	Comparison	Gemini	Absent	—
how do I track activation in a B2B product	Informational	Claude	Mentioned	no ordered list
analytics tools with EU data residency	Commercial	Google AI Overviews	Absent	—

What each row records

Mention type, kept separate: recommended, listed, mentioned in passing, or absent. Being named is not the same as being recommended, so the two are scored apart.
Position only where the answer contained an objectively ordered list. Prose answers have no rank, and inventing one would be fabrication.
The citations the engine attached to that answer, canonicalised and classified.
The execution status, which decides whether the row may enter a metric at all.

Because the raw rows are kept, any headline figure can be opened until you reach the individual answers behind it.

How the score is built

Explain

Who took the slot, and what the answer was built from.

An answer is assembled from sources. When a competitor is named and you are not, the explanation usually lives in the citation set and in the pages behind it — so that is where AnswerRank looks.

Share of voice

Your slice of the answers in your category

Every brand named across your valid checks is counted, so you see the full set that AI answers put in front of buyers — including the rivals nobody on your team named in a positioning deck.

Competitor intelligence

Citations

The domains behind the answers

Cited sources are canonicalised, deduplicated and classified as owned, competitor-owned, review platform, community, publication or third party. The sharpest signal is the source that cites your competitors and never you.

Citation intelligence

Site audit

What your own pages give an engine to work with

Access for AI crawlers, answerable structure, entity clarity, evidence a model can quote, and freshness — reported per page, with every finding attached to the URL it came from.

Site audit

Association, not causation

Findings are phrased as “evidence suggests”, “associated with” and “likely contributing factor”. AnswerRank will not tell you that a citation gap caused a loss, because answer generation is not observable from the outside.

Improve

A finding nobody can act on is not finished.

Recommendations arrive scoped, prioritised and attached to the evidence that produced them, so the person who has to do the work can see why it landed on their list.

Impact, effort and confidence on every item, so a backlog can be ordered without a meeting.
Linked evidence: the affected prompts, the competitor citations, the pages involved.
Content briefs are drafts for a human to verify and publish. AnswerRank does not push anything to your CMS.
Status moves open → in progress → implemented, or dismissed with a reason that stays on the record.

Inside the action engine

Demo dataset

Recommendation

Publish an Acme vs Northwind comparison page

Impact: HighEffort: MediumConfidence: Medium

Evidence

12 comparison prompts where Northwind is named and Acme is not
5 answers citing a competitor-owned comparison page
No owned page targeting the comparison topic

Confidence describes the strength of the observation behind the recommendation. It is not a forecast of the outcome.

Demo dataset

Experiment · comparison pageCompleted

Before

n = 64 valid

After

n = 61 valid

Evaluation window: 14 days
Prompt set: Frozen · 32 prompts

Same prompts, same engines, same market, before and after. Competitor changes, model updates and index refreshes are uncontrolled — an experiment is evidence, not proof.

Prove

Re-run the identical checks, then read the delta honestly.

An experiment compares the same prompt × engine × market set before and after a change, reports the movement with its sample size, and states the evaluation window on the face of the result.

Below 20 valid checks per side, AnswerRank refuses to declare a result and labels the deltas directional. A verdict built on a thin sample is worse than no verdict.

How experiments work

Data integrity

What is allowed to become a number.

Only two execution statuses enter a metric. Everything else means we failed to observe, and a failure to observe is not evidence of absence — so it is dropped from the numerator and the denominator alike.

Execution statuses and whether they enter metrics
Status	In metrics	Meaning
success	Valid	The engine answered and we parsed the answer.
no_answer	Valid	The engine answered and your brand genuinely was not there. A real absence, and it counts against you.
provider_error	Excluded	The provider failed. We observed nothing, so nothing is recorded.
timeout	Excluded	The request did not return in time. Retried, never counted as absence.
rate_limited	Excluded	We were throttled. A quota ceiling is not a visibility signal.
unsupported	Excluded	The engine cannot serve this prompt, market or device combination.
not_configured	Excluded	No credentials for that provider. The engine renders as unavailable, never as connected.

The practical consequence: a provider outage cannot look like a visibility collapse, and a missing API key cannot quietly render as a zero.

Coverage

Each engine, and exactly what surface we read.

Every engine card in the product renders from a single source of truth, so a claim cannot drift between this page and the dashboard.

Engine coverage, collection surface and limitations
Engine	Collection mode	Surface	Limitation
OpenAI web-grounded API	Official grounded API	OpenAI Responses API + web search tool	Reflects the web-grounded API surface. It is not a capture of a signed-in ChatGPT session, which is personalised by account history and settings we cannot observe.
Claude web-search API	Official grounded API	Anthropic Messages API + web search tool	Reflects the web-search API surface. It is not a capture of a signed-in claude.ai conversation.
Gemini + Google Search grounding	Official grounded API	Gemini API + googleSearch grounding tool	Grounding sources are those returned in grounding metadata. Consumer Gemini app results may differ.
Perplexity web-grounded API	Official grounded API	Perplexity Sonar API	Reflects the API surface. Perplexity’s consumer product applies additional personalisation and model routing.
Observed Google AI Overview SERP data	Observed SERP data	DataForSEO SERP API (ai_overview element)	AI Overviews do not appear for every query, and presence varies by location, device and time. Absence in a sample is not proof of permanent absence.

The boundary we hold

AnswerRank measures web-grounded provider APIs and licensed structured SERP data. It does not replicate a signed-in consumer chat session, which is shaped by account history, memory, subscription tier, region and model routing that no third party can observe.

We do not scrape authenticated consumer interfaces. That is a product boundary rather than a technical shortcut — and it is the reason these numbers survive scrutiny.

Go deeper on a stage

See what AI answers say about you.

Run a free visibility check on your website. No signup, no card.

No signup required for your first visibility check.

Every figure AnswerRank reports carries its sample size and how it was collected.

Read the methodology