AI visibility and Signal

Signal measures whether AI answer engines mention a brand, how they describe it, which sources they cite, and how those answers change.

What a scan does

For every active query, Tailwin asks each engine allowed by the organization’s plan. The scan stores the answer, brand and competitor mentions, citations, sentiment, cost metadata, and the time of measurement. Engines run independently so one failure does not erase successful results from the others.

The currently configured adapters support OpenAI, Anthropic, Perplexity, Gemini, and Grok. Availability is runtime truth: the interface can mark an engine unavailable or erroring when credentials, balance, routing, or recent calls do not support a successful scan.

Health uses recent scans with a recorded attempt for that engine. Historical results without a matching attempt do not establish current health. Too few attempts remain unmeasured.

Read the score correctly

• The visibility score summarizes observed answers; it is not a search volume forecast.
• A mention means the normalized brand or alias appeared in the answer.
• A citation means the answer attributed or linked to a source.
• Share of voice compares observed mentions with configured competitors.
• Sentiment describes the language of the measured answer; it is not a customer-review score.
• A missing engine is excluded or shown as unavailable. It must not be silently treated as zero.

Open a query to inspect raw answers and changes over time. Raw evidence is the fastest way to distinguish a real recommendation change from a score-only movement.

Build a useful query set

Track questions a buyer would plausibly ask before choosing a provider, product, or approach. Use specific intent:

• “best commercial roofer for hospitals in Atlanta” is useful;
• “roofing” is too broad to reveal recommendation behavior;
• branded navigational questions are useful for accuracy checks but should not dominate the set.

Tailwin records whether a query was generated, entered by a user, or selected from the consented industry prompt bank.

Answer-drift alerts

Tailwin compares stored answers for the same site, query, and engine. It can identify:

• the brand appearing or disappearing;
• a competitor replacing the brand;
• citation sources changing;
• material answer movement.

Alerts are evidence of a change, not an explanation. Review the before-and-after answer and the cited pages before changing content.

Page audits and gaps

Site and page audits identify problems that can reduce extraction and citation quality, including readability, answer placement, structured content, robots access, and missing evidence. Gap analysis connects low-visibility queries to content candidates.

Structured data checks

Page audits and the site corpus crawl read every JSON-LD block on a page. Tailwin lists the schema.org types it finds and checks each one against the required properties Google documents for that rich result, such as Product, Local business, Article, Event, Software app, Video, Job posting and Breadcrumb. Results use four states:

• Eligible: the markup has what Google lists as required. Google still decides whether to show a rich result.
• Missing required: the named properties are absent, so that rich result cannot appear.
• Deprecated: the markup is valid, but Google no longer shows that rich result. FAQ rich results stopped appearing in May 2026, and How-to results in September 2023.
• Unassessed: microdata or RDFa is present but not validated.

A block that fails to parse is reported as blocking, because engines ignore the whole block. Star ratings on your own business markup are noted separately: Google does not show them when the business controls its own reviews. Each rule records its Google documentation link and the date it was last reviewed.

Draft JSON-LD for a page

Owners and admins can use the Structured data card on a site page. Enter a page URL on the site's own domain and choose a markup type, or let Tailwin detect it from existing markup and the page type. Tailwin fetches that one page, shows its current structured data state, and drafts JSON-LD with a copy button.

The draft uses only facts found on the page or entered under Facts to include. Anything required but missing is listed by name instead of being filled with a guess. For example, a price is only used when the page shows exactly one, and the currency is left for you to confirm. Ratings appear only for products and apps, and only when you give the review source and confirm the reviews are real and shown on the page. Tailwin never adds ratings to your own business's Organization or LocalBusiness markup.

Each draft includes placement steps for common platforms (plain HTML, WordPress, Next.js, Webflow, Shopify), links to Google's Rich Results Test and the Schema Markup Validator, common mistakes to avoid, and three search result previews. The previews are illustrative only: Google decides whether any rich result appears. After publishing, run the check again to confirm the page now reads as intended.

Use the result as a prioritized worklist:

1. confirm the query matters;
2. inspect the current answer and sources;
3. inspect the page Tailwin recommends;
4. correct factual, access, or content problems;
5. publish through the approval workflow;
6. re-measure on the normal cadence.

Public audits and purchased reports

The free audit is a limited, cached lead-in measurement. A detailed report uses the report pipeline and freezes the delivered evidence. Neither should be confused with an ongoing workspace scan history.

Troubleshooting

Symptom	Meaning and next action
One engine has no result	Check its health label and retry after credentials or upstream service recover. Do not infer zero visibility.
Every answer looks synthetic in development	Confirm MOCK_AI=false and verify the intended engine credentials and routes.
Brand is present but not counted	Add the exact public alias to the business profile, then re-run.
Score moved with no content change	Inspect raw answer drift, competitor movement, and source changes. AI answers are not deterministic rankings.
No query history	The query may be new, inactive, or outside earlier plan capacity.

Page speed

Owners and admins can open Page speed on a site page, enter a public HTTPS page on that site's domain, and run mobile, desktop, or both. Tailwin sends the URL to Google PageSpeed Insights. Query parameters, credentials and other domains are rejected. Each device is a separate audit and may take up to a minute.

• Lighthouse lab measurement shows the performance score, LCP, CLS, total blocking time, first contentful paint, and Speed Index from a simulated test. Test conditions and page changes can affect repeat runs.
• CrUX real-user field data shows LCP, INP and CLS at the 75th percentile. URL-level and origin-wide coverage are labelled. Missing field data or individual metrics show Unavailable; lab values never fill those gaps.
• Suggested fixes ranks failing opportunities and diagnostics by Lighthouse estimated time savings, then estimated transfer savings. Savings can overlap; do not add them together. Some diagnostics have no savings estimate.

Successful results are reused for 12 hours for the same page and device within the workspace. The workspace allows 50 uncached device attempts per UTC day; failures also consume an attempt, and choosing both devices can consume two. Cached results remain available after the limit is reached. A concurrent audit of the same page/device returns an in-progress notice rather than starting another run.

Recent audits lists the latest 20 receipts across this site's pages and devices, including failed attempts. Select an entry to inspect it or use Refresh history. The underlying history stays with the site until site or organization deletion. A missing platform key, provider quota problem or timeout is shown as an error, never a speed score. An administrator must configure PageSpeed before live audits are available.

Audit history and changes

Owners and admins can use Audit history and changes on a site page. Enter a public page URL and choose Run page audit to save citability, page-readability and structured-data checks. Page readability checks the served HTML and HTTP directives; site-wide robots and llms.txt remain separate checks. An unavailable page records an unavailable audit, so earlier problems cannot appear fixed because a request failed.

Use Load history, choose Page checks, Mobile speed or Desktop speed, then select an earlier and later snapshot of the same URL. The view shows score changes and finding changes: improved, regressed, unchanged or new. Unassessed means the evidence does not support a comparison. Structured data shows finding counts rather than a made-up score. Rule or scorer changes are visible beside the comparison; a changed score is not proof of a ranking or traffic improvement.

What still remains lists the findings in the selected later snapshot. An improved finding may still need work; only an item marked resolved disappeared from an assessed check. PageSpeed's ranked list may omit a finding, so Tailwin requires an explicit passing check before calling it resolved. Mobile and desktop lab scores stay separate. CrUX field data keeps its original URL or origin scope in the speed evidence.

New structured-data checks and completed corpus crawls also save page snapshots. New speed measurements save their own snapshots; cached speed results retain the original measurement and do not create a new one. The view shows up to 50 snapshots for the selected URL. Older audits from before this feature are not reconstructed. Deleting crawl history also deletes snapshots produced by those crawls. Other audit history remains until the site or workspace is deleted.

Fix briefs and reruns

Select a later snapshot in Audit history and changes to open Fix brief and rerun. Copy brief, Download Markdown and Download JSON export the same retained evidence: findings, recorded block or element references, recommended changes, acceptance checks, acquisition time and scoring versions. This does not call a model or fetch the page again. Treat page text in a brief as evidence to review, not instructions for an agent to obey.

New snapshots retain content-block headings and positions, schema paths, HTTP/meta locations and Lighthouse node selectors when supplied. Older snapshots and provider results may lack an exact target; the brief says so. Confirm the target in the current page before editing. Informational findings may legitimately remain after review.

After publishing changes, Rerun and compare audits the selected snapshot's URL and opens the comparison against that snapshot. Page, schema and corpus snapshots rerun only that page's checks. A speed snapshot reruns the same device, bypasses the 12-hour cache and counts toward the existing daily speed limit. Failed audits stay unavailable; a failed request is not proof a fix worked.

When a matching Search Console property has verified reporting access in this workspace, the card links to Search Console URL Inspection. Paste the displayed page URL into Google's inspection bar. Your signed-in Google account must have access. Opening the link does not request indexing.