The State of AI Shopping Citations — 2026

The 1Digital^® Workspace AI Visibility Monitor — an internal production tool that runs a defined prompt panel through five live AI engines (OpenAI ChatGPT, Anthropic Claude, Perplexity Sonar, Google Gemini, Google AI Overviews via SerpAPI) on a weekly cadence, parses each answer for brand mentions and citations, and writes the parsed results to a Supabase-backed dashboard. This report aggregates the parsed-response table over the Q2 2026 pilot window. Full methodology here.

The Q2 2026 pilot window captured 28 successful provider responses across the 5-engine panel — a baseline reading from the Workspace's first reporting week. Subsequent reports will add weekly increments; the pilot is now in continuous operation and the next report will cover the 12-week trailing window with materially larger N. The methodology, prompt set, and engine list are stable; the panel grows with each weekly cron.

Because the engine-to-engine spread is so wide that even a tightly-bounded baseline reading is meaningful. The 86-point gap between Perplexity and Google AI Overviews on mention rate, and the 0% AI-Overviews result on a prompt set where every Overview triggered, are findings that scale won't reverse — they reflect each engine's posture toward brand surfacing, not sampling noise. Larger-N reports will refine the magnitudes; the directional findings already hold.

The Workspace tracks both client cohorts and 1Digital^®'s own AI surface as a calibration baseline — so 1Digital is named openly in the agency-comparison section below. Client-cohort tables remain private and are aggregated only at the engine and panel level. This report exposes only categorical, provider-level, and aggregate-of-N findings — the engines, the citation/mention rates per engine, cross-engine overlap counts, and discovered-brand counts (numbers only, never names) — plus the explicit 1Digital-vs-peer-agency comparison computed against actively-monitored agency competitors. No client name and no specific competitor brand appears anywhere on this page or in any downstream artifact.

Mention rate is the % of successful answers in which the brand name (or a registered alias) appears in the answer text, regardless of whether the engine sourced it. Citation rate is the % in which a URL from the brand's domain set appears in the engine's source list. Citation is the stronger, more attributable signal — it predicts click-through. Mention is the broader signal — it captures the paraphrase-without-credit cases that still drive brand recall. Engines vary in their citation density: Perplexity cites heavily; OpenAI mentions in narrative without sourcing. Both metrics matter; neither tells the whole story alone.

Because in the Q2 2026 pilot window, AI Overviews behaved as a category-answer surface, not a brand-recommendation surface. Every pilot prompt triggered an Overview (no no_ai_overview statuses in the data), and the returned Overviews answered the category question in generic terms — feature lists, general guidance, decision criteria — without naming specific brands. This is consistent with Google's public framing of AI Overviews as an evolution of featured snippets. The optimization play for AI Overviews is owning the category-defining answer block, not chasing per-brand citations inside it.

Claude's web-search invocation timed out on a majority of pilot calls in this window — a known operational variable on the Anthropic web-search beta. The Workspace logs the failure mode separately (status=error, with the timeout captured), and Claude's mention rate is computed only against successful responses, not against the timed-out calls. The elevated error rate is a finding in itself: brands optimizing for cross-engine surface need to account for engine-side reliability differences when reading week-over-week trends, and dashboards that fold errors into the denominator will misread engine outages as citation drops.

This report is the public-facing read of the methodology described at /methodology/citation-share-monitoring. Inside client engagements, the same Workspace runs a per-engagement prompt panel — tracked-keyword variations, category-defining synthesis prompts, competitor-set prompts — and produces a per-client dashboard with week-over-week movement, content-brief generation on drops, and re-measurement after content ships. The aggregate-public view (this report) and the per-engagement view (clients' dashboards) share the same monitor; only the prompt set and subject list change.

The pilot is in continuous weekly operation. The next report (Q3 2026, published August) will cover the 12-week trailing window — materially larger N, week-over-week trend lines per engine, intent-bucket breakdowns (informational / commercial / comparison) with statistical power, and the first cross-vertical comparisons as additional industry panels come online. The methodology is stable; only the data window grows.

The Workspace data is private — it contains client-confidential prompt sets and monitored brand lists. The report aggregates and findings published here are the public surface. If you want a citation-share read on your own brand or a category we don't currently monitor, the fastest path is a scoped AEO engagement — we'll build your prompt panel, run it through the same five engines, and report your numbers against the same definitions.