How to Measure AI Referrals for Mobile App Acquisition

Measure each stage of an AI-assisted app journey on its own terms: a crawler reads context, an assistant represents the app, and a person may follow a link into the experience. Instrument the path from technical discovery to a qualified in-app outcome while keeping fetches, recommendations and downstream actions distinct.

Use six practical checkpoints: access, retrieval, recommendations, routing, in-app outcomes and map maintenance. Blue Jetty records app-map requests; your existing analytics and MMP provide the downstream observations available in your setup.

Stage 1: Technical discovery

Start with server logs and sitemap tooling. Record requests to robots, sitemaps, app profiles, entity pages, feeds, and API descriptions. Group known user agents by documented purpose rather than combining every “AI bot” label.

OpenAI’s publisher FAQ identifies OAI-SearchBot for ChatGPT search visibility and says ChatGPT referral URLs include utm_source=chatgpt.com. Anthropic documents separate crawler roles as well. Preserve the user agent, path, status, response bytes, and timestamp, but do not assume a claimed user-agent string proves identity.

Success at this stage means important public pages returned useful content without a policy or infrastructure block. It does not mean indexing.

Stage 2: Indexing and retrieval

Use webmaster tools where available to measure sitemap processing, indexed pages, impressions, and query classes. Run a fixed evaluation set in selected answer products with browsing enabled, recording the date, product/model label, prompt, response, citations, and region.

Include brand, category, constraint and comparison questions. Test whether the assistant can explain a relevant app experience and find the publisher's preferred link. Keep the sample consistent enough to compare changes over time.

Stage 3: Citation and referral

Track whether the assistant names the application, cites the correct source, and links to the intended canonical page. Referral analytics should capture known source parameters and referrers while respecting consent and privacy settings.

Add a destination event before leaving the web property. Store the selected entity ID, destination class, platform decision, and anonymous experiment identifier—not the full user prompt or sensitive profile data.

Stage 4: Routing quality

Measure what happens after the click:

• entity-level route rate;
• app-open versus web-fallback rate;
• broken or redirected destination rate;
• authentication interruption;
• regional unavailability;
• loss of selected context during install.

Universal Links and Android App Links make intent-preserving handoff possible, but only device-level testing shows whether a particular path works.

Stage 5: Qualified outcomes

An install is not always the right success event. A user might already have the app. Choose a downstream action aligned with the discovered entity: class started, reservation viewed, route saved, product inspected, or provider contacted.

Use aggregate attribution when user-level linkage would be invasive or unreliable. Compare AI-referred cohorts with other discovery channels, but account for small samples and brand demand.

Stage 6: Map maintenance

Track the publisher work that keeps the representation accurate: active revision, website connection, authorization state, stale destinations, expired offers and time since the latest successful update. Use those signals to repair the app map alongside releases rather than growing the record count for its own sake.

Blue Jetty gives app teams one place to maintain experiences, current offers and preferred links in a machine-readable app map. AI assistants can retrieve that context, while your team evaluates recommendations and downstream activity through the appropriate reporting tools.

Create a baseline before outreach. Review the funnel alongside map freshness and real-device checks. Use publisher and MMP reporting for downstream outcomes, and keep crawler activity as its own clearly labeled observation.