What you get in an AI Visibility check

Part 1

What AI says about you

Does AI name us, and how does it describe us?

Start with the prominence index, then look at how often your brand is named and the tone of those answers.

Prominence and mentions

The index out of 100 weights where the brand appears in usable answers. The mention count shows how many of those answers name it.

Tone when named

See the share of brand-naming answers that are positive, with the underlying count. A prominent answer can also be critical.

Descriptions and recurring themes

Read the summary, repeated phrases and what assistants praise or criticize. Where there is an earlier result in the series, look for changes in wording.

Category questions

Questions that leave your name out have a separate reading. They show when assistants bring up your brand on their own.

Ask Polly

Ask questions about this finished check in the chat at the bottom right. Polly answers from its saved research; open the evidence before acting on an answer.

Part 2

Where you stand, question by question

Which questions and assistants make the difference?

The matrix has your exact questions down one side and the assistants across the top. Each answered cell shows how the brand appeared in that reply.

Read a cell

Prominent mention means a main subject or first option. Listed means one of the options. In passing means a brief mention. Absent means the assistant gave a usable answer and left your brand out.

Open the evidence

Read the question, the saved response and any citations the assistant provided. The per-assistant tables let you inspect answers one at a time.

Check coverage

An unanswered question and a search with no Google AI Overview have their own labels. Both stay out of answer scores, so the usable-answer count can differ by assistant.

Follow the buyer journey

See how the answers differ across discovery, comparison and other buying stages. Topics and intent help you find the questions that matter to your audience.

Your buyer labels

Name the buyer behind each question and group the saved answers by buyer or tag. Labels organize existing results and carry into later checks; they do not ask the assistants again.

Understand variation

Where a category question was asked more than once, repeated samples show how the replies varied. A single cell is one collected observation.

Part 3

Who wins the answer

Which rivals appear, where do they appear, and what is said about them?

The check puts your brand and the rivals it found in the same set of answers.

Share of voice

Compare the number of usable answers that name each brand. Several brands can appear in one answer, so their shares can overlap.

Ordering and perception

See which names appear first, which names assistants reach for when you are absent, and the tone and themes attached to each rival where there is evidence.

Question-by-question comparison

Open side-by-side evidence to see when another brand appears and which pages supported that answer. A separate rival check has its own scope and order controls.

Part 4

What the assistants read

Which pages support what the assistants say?

Open the saved pages and citations behind the check and see which ones appeared across several answers.

Retrieved pages and cited sites

Pages reported as retrieved and links cited in an answer are shown separately. An assistant can consult a page without linking to it in the reply.

Who owns the sources

Compare your own website and official profiles with rival and independent sources. See which kinds of pages appear, such as reviews, media or product pages.

Sections of your site

See which parts of your own website the assistants read, grouped by the first part of each page's address.

Source quality and coverage

Inspect source dates and market fit where they can be established, see missing retrieval information, and check which sources recur in negative or rival-favoring answers.

Citation gaps and broken links

Find pages feeding answers that name rivals without naming you, and broken links assistants send people to when the check can verify them.

Part 5

What to do next

Which change should we make first?

Actions are tied to the questions, claims, sources and website findings in this check.

Prioritized actions

Start with the issues backed by the clearest evidence. Each action explains the finding and takes you to the relevant part of the result.

Claim evidence and possible conflicts

Inspect exact excerpts selected throughout answers, their available citations and dates, and possible conflicting values or negation. A few unresolved claims can be reviewed against discovered official pages. Related text does not establish truth.

Claim review and correction

Review statements the assistants made as fact, add an approved fact with its source, and prepare a correction using the evidence you have.

Website checks and publishing briefs

Review your site checklist, check an individual page and see briefs for questions your pages could answer better. Page tools depend on accessible pages.

Approved facts and saved voice

Save your approved brand facts and writing voice in the Brand hub. Drafts and rewrite advice are instructed to follow them; review the wording and sources before publishing.

Fill in and remember

Replace a draft's missing fact once. You can save it with a public source as an approved brand fact for later drafts.

Follow-up work

Save an action with an owner or due date. Keep the baseline so a comparable check can show what changed later.

Part 6

Your research workspace

How do we use the check after reading it?

The private Brand Dashboard keeps this check with the other research for the same brand and market.

Filters and saved views

Open the same market, question set, assistant or date again. Review source classifications and flags without rewriting the original results.

Comparable changes

A later check can compare matching questions and assistants, explain gaps in coverage, and show changes in answers and sources.

Website outcomes

Bring in analytics, search or crawler data to review visits, conversions and crawler activity. Your pages, one row each joins the available evidence by page; missing measurements stay empty.

Crawler identity

A server log can distinguish published crawler addresses from visits using only a crawler's name, when the log contains the visitors' real addresses. Visitor addresses are processed in your browser.

Site watch

Switch on a daily look at up to 20 pages of the checked website and receive alerts for important changes. It uses no credit, respects robots.txt and pauses 90 days after the brand's newest check.

Tell search engines about a change

Verify a key file on the website, then announce changed pages through IndexNow. Submission does not guarantee indexing or an AI recommendation.

Client and team delivery

Invite a colleague or client to one brand and market, read only. They sign in with the invited address and see the latest shared results without private answers, downloads or editing controls. You can remove access.

What is included and what needs setup

In one check

Questions are asked live across twelve assistants and search experiences. The set covers ChatGPT, Claude, Grok, Gemini, Perplexity, Mistral, DeepSeek, Kimi, Microsoft Copilot, Brave Search, Google AI Overviews and Google AI Mode. Each result reflects the responses and source data available on that run.

When you add another result or website data

Change comparisons need earlier research in the same scope. Linked AI-versus-consumer examples need a consumer report for the same brand, market and language from the preceding 30 days. Website outcomes need your analytics, search or crawler data.

Separate options

Extra questions, city research and daily tracking are paid options. Recurring full checks use the plan allowance or a credit when they run. Review the price and scope before enabling an option.

Delivery and downloads

Your first result lands within about an hour. Any question an assistant fails to answer in time is re-asked automatically 15 minutes, 3 hours and 9 hours later, at no extra cost, and we email you when the check is complete.

• Private result and full PDF

  Review the interactive answers and sources in PollyX or download the check for reading and sharing.

• Full deck, executive slides and pitch PDF

  Choose the full PowerPoint, a 10-slide executive preset or a short pitch PDF.

• Separate Answers and Sources CSVs

  Use the response-level table to audit questions and the source table to examine the evidence. Markdown supports document workflows.

Open Export files on the finished result. The brand workspace also lets you prepare client files and manage sharing. Scheduled research and scheduled delivery have separate controls.

How to read the result fairly

• A prominence score describes the brand's position in the answers collected. Read tone and category questions separately to understand praise and discovery.
• An answer can vary from one request to the next. Review the sample and coverage before treating an observed change as a trend.
• The source panel distinguishes reported retrieval, citations and unknown retrieval. Links and dates reflect what the assistants and pages made available.