Guide

Constants & sources

Every hardcoded number in the traffic and revenue math — with its source, uncertainty band, and last review date.

Registry constants@v1.2026-07

21 constants tracked · 7 flagged editorial (no primary source; defensible default). Everything else is anchored to an external panel, first-party OTel data, or a published disclosure.

Calibration accuracy for these constants is published live at /trust/accuracy.

Intent → AI shift

Share of Google-equivalent volume that has migrated into AI assistants, per intent.

Constant	Value	Uncertainty	Source	Reviewed
AI share of research queries `shift.research` Research is the highest-AI-shifted intent after coding.	32%	±10pp (log-uniform)	Similarweb assistant panel + a16z LLM-usage report + OpenAI/Perplexity disclosures, blended May 2026 `src/lib/agent-volume-constants.ts:INTENT_AI_SHIFT`	2026-05
AI share of comparison queries `shift.comparison`	28%	±10pp	Same blend as research; comparison queries slightly lag research shift. `src/lib/agent-volume-constants.ts:INTENT_AI_SHIFT`	2026-05
AI share of how-to queries `shift.howto`	25%	±10pp	Similarweb + Datos panel, May 2026 `src/lib/agent-volume-constants.ts:INTENT_AI_SHIFT`	2026-05
AI share of coding queries `shift.coding`	45%	±12pp	GitHub Octoverse 2025 + Stack Overflow Developer Survey 2025 (LLM-usage question) `src/lib/agent-volume-constants.ts:INTENT_AI_SHIFT`	2026-05
AI share of shopping queries `shift.shopping`	10%	±5pp	Adobe Analytics AI-referred retail traffic + Similarweb, calibrated May 2026 `src/lib/agent-volume-constants.ts:INTENT_AI_SHIFT`	2026-05
AI share of transactional queries `shift.transactional`editorial	12%	±6pp	Editorial blend — transactional is under-instrumented in public panels. `src/lib/agent-volume-constants.ts:INTENT_AI_SHIFT`	2026-05
AI share of navigational queries `shift.navigational`	3%	±2pp	Navigational is bookmark/URL-bar dominated; low AI displacement. `src/lib/agent-volume-constants.ts:INTENT_AI_SHIFT`	2026-05
AI share of local queries `shift.local`	4%	±3pp	Local still overwhelmingly Google/Apple Maps; AI displacement lags. `src/lib/agent-volume-constants.ts:INTENT_AI_SHIFT`	2026-05
AI share of general queries (fallback) `shift.general`editorial	12%	±8pp	Editorial fallback for unclassified intents. `src/lib/agent-volume-constants.ts:INTENT_AI_SHIFT`	2026-05

Intent → agent shares

How assistant traffic splits across agents, conditioned on the intent type.

Constant	Value	Uncertainty	Source	Reviewed
Per-agent share of assistant traffic conditioned on intent `shares.overview` Each row sums to 1.0. Apple Intelligence is zero except for local/navigational — it is a personal-context surface, not a discovery agent. Meta AI is intent-scoped to shopping/local/comparison/navigational.	See INTENT_AGENT_SHARES (9 intents × 8 agents = 72 cells)	±25% relative per cell (higher on Meta AI / Apple Intelligence rows)	Similarweb monthly assistant traffic + OpenAI/Anthropic/Perplexity/xAI disclosures, per-intent split from Datos panel. May 2026 snapshot. `src/lib/agent-volume-constants.ts:INTENT_AGENT_SHARES`	2026-05

Agent-answer CTR curve

Click-through to the brand's destination by answer position.

Constant	Value	Uncertainty	Source	Reviewed
Agent-answer click-through at rank 1 `ctr.rank1`	30%	±8pp	First-party OTel click logs from instrumented brands (n=41, Q1 2026) — agent answer position → destination CTR. `src/lib/revenue-estimator.ts:AGENT_CTR_BY_RANK`	2026-04
Agent-answer CTR at ranks 2–5 `ctr.rank2to5`	18% / 10% / 6% / 4%	±5pp each	Same OTel click panel. Curve is steeper than Google SERP CTR — agents surface fewer results. `src/lib/revenue-estimator.ts:AGENT_CTR_BY_RANK`	2026-04
Agent-answer CTR at ranks 6–10 `ctr.rank6to10`editorial	2.5% / 2.0% / 1.8% / 1.5% / 1.2%	±1pp each	Extrapolated from ranks 1–5 with a power-law fit; below rank 5 the panel is thin. `src/lib/revenue-estimator.ts:AGENT_CTR_BY_RANK`	2026-04
Passing-mention CTR (no rank) `ctr.unranked`	0.5%	±0.3pp	OTel click panel — brand mentioned but not surfaced as an option. `src/lib/revenue-estimator.ts:AGENT_CTR_UNRANKED`	2026-04

Conversion rates

Default agent-session conversion rates. Overridden per-workspace when analytics is connected.

Constant	Value	Uncertainty	Source	Reviewed
Default agent-session conversion rate `conv.default` Overridden per-workspace by measured GA4/Segment conversion when available.	1.0% – 3.0%	Band is the interval, not a point estimate	Adobe Digital Economy Index 2025 (AI-referred conversion is 9% above baseline direct traffic; baseline retail conv is ~2%). `src/lib/revenue-estimator.ts:DEFAULTS.convPctLow/High`	2026-04

Average order value

Default AOV band used until the workspace supplies its own.

Constant	Value	Uncertainty	Source	Reviewed
Default average order value (USD) `aov.default`editorial Users should override with their real AOV in workspace settings to tighten the band.	$30 – $120	Band is the interval; per-category AOV overrides available in workspace settings.	Shopify Plus benchmark 2025 (SMB retail median AOV $47; SaaS trial-to-paid ARPU $80–$140). `src/lib/revenue-estimator.ts:DEFAULTS.aovUsdLow/High`	2026-04

Baseline traffic

Fallback traffic band when no data connector is linked.

Constant	Value	Uncertainty	Source	Reviewed
Baseline monthly agent visits when only audit scores are available `traffic.baseline`editorial Only used when the brand has no external data connectors linked. UI badges the estimate 'coarse'.	800 – 4,000 / mo	Wide by design — becomes irrelevant as soon as Semrush or GSC is connected.	Editorial fallback derived from the p25/p75 of instrumented brands' monthly agent visits. `src/lib/revenue-estimator.ts:DEFAULTS.baselineMonthlyAgentVisits*`	2026-04

Causal-lift model

OLS + bootstrap parameters powering the correlated-tier lift estimate.

Constant	Value	Uncertainty	Source	Reviewed
Visibility-score neutral threshold for lift attribution `lift.exposure-threshold`editorial Lift = OLS slope × max(avgVisibility − 50, 0) × 30 days. Below 50 the model returns 0 rather than negative lift.	50 (index points)	Chosen as the panel median; sensitivity ±10 changes monthly lift by <8%.	Median AIAO Index across all tracked brands, Q1 2026 (n=1,204). `src/lib/causal-lift.ts:estimateLift`	2026-04
Bootstrap replicates and CI band `lift.bootstrap-ci`	200 replicates, p10–p90 band	Band widens naturally as Pearson r weakens.	Standard bootstrap; 200 is enough for a coarse 80% CI without slowing server functions. `src/lib/causal-lift.ts:estimateLift`	2026-04

Confidence bands

Log-symmetric multipliers used to turn a modeled point estimate into a range.

Constant	Value	Uncertainty	Source	Reviewed
Confidence-band multipliers (high/medium/low) `confidence.factors`	×1.3 / ×2.0 / ×3.0 log-symmetric	Calibrated against holdout brands so that ~80% of measurements land in the medium band.	Internal calibration against the holdout panel (see /trust/accuracy). `src/lib/agent-volume-constants.ts:bandFromConfidence`	2026-07
Spread-to-confidence threshold `confidence.spread-threshold`editorial	max/min < 1.5 → High, else Medium	Coarse threshold; deliberately conservative.	Editorial rule — the log-symmetric band already absorbs the rest of the uncertainty. `src/lib/agent-volume-constants.ts:confidenceFromSpread`	2026-04

How to challenge a constant

If you have a primary source that contradicts one of these numbers, open an issue or email methodology@appkitekt.com with the constantidand the source. We re-review every constant at least quarterly and on any credible challenge.