The research-phase query has left your website. When a customer types "best certified pre-owned SUV near me" into Google, Perplexity, or ChatGPT Search, the answer arrives before your VDP does. Some dealers get cited. Most do not. And most dealer groups have zero instrumentation to know which side of that line they are on.
This playbook is the instrumentation plan for a 10-30 rooftop dealer group. It covers what to probe, how to store the data, and how to feed the signal back into content strategy. It is not a strategy for getting cited — that is a separate discipline, driven by structured content and canonical claims. This is the measurement layer that tells you whether the strategy is working.
Step 1. Map the queries that land in AI answer engines
Start with a query set — 200-500 phrases split into three buckets:
- Category intent — "best used SUV under $25k in [metro]", "certified pre-owned Toyota vs Honda", "cheapest new EV lease [state]". These are the queries AI Overviews replaces most aggressively.
- Model + rooftop — "[rooftop name] used inventory", "[rooftop name] Camry lease", "[rooftop] service loaner". The dealer-branded tail your group already ranks for.
- Question shape — "how do I finance a used car with 620 credit", "what does dealer prep fee include", "when is the best time of year to buy a truck". Perplexity and ChatGPT Search cite question-shaped content disproportionately.
Do not try to cover every possible query. 200 well-chosen phrases across your operating markets give you a stable enough signal to detect week-over-week changes.
Step 2. Build the citation-probe cron
For each query, hit each engine on a schedule. AI Overviews and Perplexity respond to their public search endpoints; ChatGPT Search has a lightweight API. Store one row per (query, engine, run_at) with the ordered list of cited URLs.
Two rules that save weeks of pain later:
- Never key on markup surrounding the citation. Google changes AI Overviews DOM every two-to-three weeks. Extract the URL list from the citation card and ignore everything else.
- Retry on empty responses before logging. Both AI Overviews and Perplexity have a ~2% empty-response rate that is not a real "no citation" — it is a transient. One empty run per query is signal-worthless; two consecutive runs is real absence.
Step 3. Store citations in a query-partitioned table
You want to be able to answer three questions in a SQL query:
- Which of my rooftops are getting cited this week, and at what rank?
- Which queries are the largest opportunity — high intent, no group citation?
- Which content of mine, when cited, holds rank across weeks?
The schema that answers all three: citation_event(id, engine, query, cited_url, rank, cited_at, run_id, rooftop_id NULLABLE). Rooftop-id is populated by a nightly enrichment job that maps cited_url to your inventory feed. Never populate it inline — the extractor should be dumb.
Step 4. Alert on new + lost citations
Two alerts, both delivered to the marketing director's inbox on Monday morning:
- New citation — a rooftop URL that has never been cited before, appears in a top-3 position on any engine. This is a strategy signal: understand why and repeat it.
- Lost citation — a URL that has been cited for 4+ consecutive weeks disappears from the top-5. This is a diagnostic signal: something regressed. Usually a page update that broke the schema or introduced ambiguity.
Do not alert on rank shifts within top-5. Rank noise is high and the false-positive rate is unmanageable.
Step 5. Feed the signal into content strategy
The instrumentation is worthless if the output does not change what gets written. Every quarter, the marketing director should be able to answer: "Which of the 200 tracked queries have we moved into a top-3 citation, and what did we do to move them?" If that answer is empty two quarters in a row, the strategy is not working and the query set or the content approach needs to change.
Failure modes we have seen
- Overprobing. Groups run every query every day, burn quota, and then throttle back to weekly. Start at weekly Wednesday + Sunday. Only tighten if you catch a real signal you would have missed.
- Rooftop-level noise. Small rooftops with thin content register huge percentage swings from noise. Aggregate to group level for trend detection; use rooftop-level data only for diagnosis.
- Storing only the "current" citation. Overwriting citation state on each run destroys the historical signal. Never overwrite — append. Storage is cheap; a bad decision made from a missing history is expensive.
What to build first
If you have one week: build the probe for AI Overviews across your top 50 queries, store the citation events, and have the marketing director look at the raw data by hand every Monday. No dashboard, no alerts. Just eyes on the data. Two weeks of that will teach you more than a quarter of dashboarding.
How AUTONOMi Runs This Playbook For You
Every step above — the probe cron, the citation table, the delta report, the weekly review — is what ECHO, AUTONOMi's answer-engine visibility layer, does end-to-end. The probes run on the twice-a-week Wednesday/Sunday cadence recommended above, against Perplexity, ChatGPT Search, and Google AI Overviews simultaneously. Every citation is stored in the LLMO Claim Graph — a hash-chained ledger of which URL was cited for which query on which date — so you can prove citation gain to a dealer principal without asking them to trust a screenshot.
When a rooftop loses a citation, ECHO writes a real Content Gap ticket to the rooftop's marketing manager with the exact query, the winning URL, and a proposed content change. When a rooftop gains one, AUTONOMi rewrites the winning content to lock the citation in. No spreadsheet, no manual scoring, no 60-page Perplexity export to parse. The whole ledger is audit-ready because every write goes through AXIOM, AUTONOMi's governance layer — same integrity guarantee as the attribution cadence article.
The instrumentation described here takes an internal team six to twelve weeks to build once you have the right operator; AUTONOMi turns it on the day you sign up. If the tradeoff is your marketing lead spending Q3 on a citation-probe cron versus running the actual sales calendar, start now — or talk to us about a group deployment if you're a 5+ rooftop group and want the citation report at the group level from day one.
The instrumentation is not the strategy. But without it, you are optimizing blind.
