Limy AI Glossary

The Limy AI Glossary provides clear, concise definitions of key terms used across the agentic web ecosystem. As AI search engines, LLM-driven interactions, and autonomous agents reshape how users disc

Core concepts

• Prompt The user’s query to an AI surface (e.g., “best running shoes for knee pain”).

• Response The AI’s response to a prompt, which may contain links to your site.

• Prompt Hash A stable ID (hash) representing a canonicalized prompt text used for grouping/analytics.

• Answer ID A unique identifier for a specific AI answer snapshot.

• Rank/Position The placement of your link within the AI answer (e.g., first link shown).

Acquisition & tracking

• AI Impression Recorded when a tracked prompt/answer containing (or likely leading to) your link is shown.

• AI Click A click from the AI answer to your site (ideally via a decorated/redirected link).

• Redirector / Link Decorator A shortlink service that wraps destination URLs to attach IDs/UTMs for deterministic attribution.

• Destination URL (Dest URL) The specific page on your site that receives the AI-driven visit.

• UTM Parameters Standard marketing tags (e.g., utm_source=chatgpt) persisted on arrival for analytics.

• AI Params Non-UTM tags used by Limy (e.g., answer_id, prompt_hash, link_id, ai_model).

• Referrer The page or service that led the user to your site (often the redirector domain).

• Session A continuous period of user activity on your site.

• Sessionization Logic that groups web events into sessions (server-side preferred).

• Client ID (CID) First-party identifier stored in a cookie/local storage and joined to sessions.

• Server Session ID A server-generated session key used to stitch events and reduce reliance on client cookies.

• Event Collector (SDK/Edge) Limy’s server-side endpoint or edge function that receives web/app events.

• Limy Pixel The lightweight snippet/SDK that initializes IDs and forwards events server-side.

On-site events & commerce

• Page View A view of any page; typically the first event after AI Click.

• Key Event Milestone interactions (e.g., view_item, add_to_cart, begin_checkout, lead_submit).

• Purchase Completed order with order_id, value, currency, items.

• Order Value / Revenue Monetary value attributed to a purchase after discounts but before refunds.

• Conversion (CVR) A desired outcome (often purchase or qualified lead) divided by a base (clicks or sessions).

Identity & stitching

• Identity Graph A table joining client_id, server_session_id, hashed email, and user/account IDs.

• Deterministic Match High-certainty connection (e.g., redirector link_id + session, or logged-in user).

• Semi-Deterministic Match Strong but not perfect signals (e.g., redirector referrer + near-instant session).

• Probabilistic Match Statistical linkage using features like time gap, geo, device family, and URL similarity.

• Cross-Device Stitching Merging identities across devices via hashed email, login, or server-side IDs.

Attribution & models

• Attribution The method of assigning credit (and revenue) to touchpoints that influenced a conversion.

• Attribution Model (Selector) Switchable logic: Last-Touch, First-Touch, Position-Based (40/40/20), Time-Decay, or Data-Driven.

• Last-Touch 100% credit to the final touchpoint before conversion.

• First-Touch 100% credit to the initial AI touchpoint (first impression/click).

• Position-Based (40/40/20) 40% to first AI touch, 40% to last pre-purchase touch, 20% split across middle touches.

• Time-Decay Credit weighted by recency (configurable half-life).

• Data-Driven Attribution Algorithmic approach (e.g., Markov/Removal Effect or Shapley) estimating each touchpoint’s incremental impact.

• Lookback Window Time span during which touches are eligible to receive credit (e.g., 7 days for clicks).

• Attribution Confidence A 0–100 score reflecting certainty of the AI→session→order linkage.

• Confidence Tiers High (≥0.80), Medium (0.60–0.79), Low (<0.60); used for reporting filters.

• Attributed Revenue Revenue portion assigned to AI touchpoints under the active model and confidence threshold.

Analytics objects

• Journey / Path The ordered set of AI and on-site events leading to a conversion.

• Journey ID A unique ID for a specific path instance used in deep links and audit trails.

• Timeline UI visualization of a journey’s events and timestamps.

• Funnel Aggregated steps from Prompt Impressions → AI Clicks → Sessions → Key Events → Purchases.

• Category A thematic grouping of prompts (e.g., “Running”, “Accessories”), used for reporting and drill-downs.

• Prompt Leaderboard A ranked table of prompts by revenue, conversions, or efficiency.

• Top URLs Report listing destination pages receiving the most AI-driven traffic/conversions.

• Anomaly A statistically significant spike or drop (e.g., day-over-day or week-over-week) with a likely cause (traffic vs CR).

• Lift Performance change versus baseline/previous period (e.g., +12%).

• KPIs Key metrics surfaced at the top of dashboards (e.g., Attributed Revenue, Conversions, Avg Confidence).

Data model & fields

• ai_impressions Table of prompt/answer exposures with prompt_hash, answer_id, model, position, time, URLs.

• ai_clicks Click rows with link_id, prompt_hash, answer_id, click time, dest URL.

• web_sessions Session metadata (first/last seen, geo, device family, first AI params).

• web_events All tracked events with event_name and properties JSON.

• orders Commerce records with value, currency, items, and session link.

• attribution_results Computed credit per order by model, with confidence and explanations.

• link_id Unique ID assigned by the redirector when generating decorated links.

• ai_model The assistant model name (e.g., “gpt-4o”) stored with impressions/clicks.

• utm_source / utm_medium / utm_campaign / utm_content Standard acquisition tags used alongside AI params.

Governance, privacy & reliability

• Consent Gate Logic that defers non-essential tracking until user consent is granted.

• Data Retention How long raw versus aggregated data are stored (e.g., 25 months for raw).

• Access Controls Role-based permissions for sensitive views (identity, raw payloads).

• Sampling Processing a subset of events for speed/cost, with safeguards for accuracy.

• Backfill Recomputing histories after schema/model changes or late-arriving data.

• Deduplication Process to prevent multiple event records for the same user action.

• Schema Versioning Version tags on event payloads/tables to manage change safely.

UI & workflow terms

• Model Picker Control to switch the active attribution model across widgets.

• Confidence Threshold Filter to include/exclude low-certainty attributions from totals.

• Drill-Down Navigation from aggregates (category) to specifics (prompts/URLs), then to journeys.

• Export CSV/JSON/PNG outputs for tables and charts.

• Virtualization Rendering technique that keeps tables fast with 1,200+ prompts.

• Tooltip Copy Short definition text used in the UI (often derived from this glossary).