See how AI answer engines talk about you — before it costs you leads

You can buy the tooling, yes — the platforms most agencies white-label run a few hundred to a couple thousand a month. What you can't buy off the shelf is the part that makes the numbers mean anything: a prompt library built from your actual buyers, sound multi-run methodology, a competitor set derived from real co-mention data, and a senior analyst who tells you whether a swing is a real change or just model randomness. A raw dashboard with no interpretation is data delivery, not insight. That's exactly the commoditised trap we position against.

Because one manual run is noise. LLMs are probabilistic — the same prompt produces a range of valid answers, and results also shift by geography, personalisation, and whether you're logged in. A serious read requires running each prompt multiple times and averaging, covering multiple engines that diverge sharply from each other, benchmarking against competitors, and tracking the trend over weeks. A single check on a single engine tells you almost nothing you can act on.

No, and anyone who promises that is lying. These are probabilistic systems — there's no fixed rank to own the way there is in classic search. The honest goal is consistent inclusion, measured as share-of-voice across a stable library of buying-intent prompts. We'd rather set that expectation correctly than sell you a number we can't control.

We run your prompt library across each engine repeatedly and average the results, because a single-run snapshot mistakes model randomness for real change. We control for geography, personalisation, and logged-in versus anonymous state rather than pretending one run is truth. We derive your competitor set from the brands the AI co-mentions with you, not a list you hand us. And we track for at least 30 days before drawing conclusions. Everything rolls up into share-of-voice, citation share, sentiment, and drift — live in Aphra, per engine.

Enough that your share-of-voice is statistically stable, weighted toward buying intent. Twenty loose prompts swing wildly and mislead leadership; brute-forcing tens of thousands adds cost without adding representativeness. We build the library from your sales and support reality, map it across awareness to conversion, validate it against real search volume, and refine it as the data reveals gaps. We deliberately track far more unbranded than branded prompts — branded prompts inflate your numbers and teach you nothing about category visibility.

ChatGPT, Perplexity, Google AI Overviews and AI Mode, Gemini, and Microsoft Copilot — tracked separately, never blended into one score. But coverage should follow where your audience actually is, not a vanity platform count. A B2B audience skews Perplexity and ChatGPT; a consumer brand skews AI Overviews and Gemini. We tune the engine mix to your buyers rather than padding a number.

Tracking is the measurement half; optimization is the doing half — the instrument panel versus the engine. This retainer tells you exactly where you stand and alerts you when it moves. AI Search Optimization (GEO) is the work that improves those numbers: answer-first content, entity and schema hardening, off-site citation building. Many clients start with tracking to get an honest baseline, then layer optimization on top. If we run both, we don't double-charge for the shared foundation.

This is the hardest and most honest question in the category, and we won't hand-wave it. AI-assisted discovery has no user-level attribution the way paid search does. So we triangulate rather than claim one clean number: share-of-voice and citation trends over time, self-reported 'how did you hear about us' data, geo or holdout lift tests where feasible, and marketing-mix modelling with an AI variable for larger accounts. We report confidence intervals and are clear about what's directional versus verified. A single-method ROI claim with no confidence interval is a red flag — from us or anyone else.

Plan on 45–60 days to a stable baseline you can actually act on — the first weeks go into building the library, establishing per-engine baselines, and running enough cycles that share-of-voice settles instead of swinging. Trend lines and drift signals get sharper from there. And AI visibility decays without ongoing work as competitors optimise and engines favour fresher sources, which is exactly why this is an ongoing retainer rather than a one-off snapshot.

AI does the grunt work it's good at: running the library across every engine repeatedly, extracting citations and mentions, classifying sentiment at scale, and watching for drift around the clock. A senior analyst owns everything that needs judgment — building the prompt library, deciding whether a move is signal or volatility, deriving the real competitor set, and writing the recommendations tied to your funnel. Nothing gets reported to you as a finding without a senior having read it first. It's the opposite of an automated PDF with no analysis attached.