Your buyers stopped clicking. They started asking.
AI answer engines now field the questions your pipeline used to come from. When the model builds its answer and your competitors get cited while you don’t, no ranking report will ever show you the deal you lost.
Tested across the engines your buyers actually ask
ChatGPT
Perplexity
Gemini
Claude
Copilot
Grok
Meta AI
7
answer engines tested
40+
buyer questions per audit
90
day first checkpoint
5
TRACE dimensions scored
An AI answer is built in three moves. You can lose at any of them.
There’s no mystery here and no magic. When someone asks an engine a buying question, a specific pipeline runs. Knowing where you fall out of that pipeline is the entire game.
MOVE 01 · RETRIEVAL
The engine pulls sources
Before writing a word, the engine retrieves a working set of pages, usually somewhere between 10 and 40. If nothing about you is in that set, the answer is over before it starts. Ranking on page one doesn’t guarantee retrieval. Different index, different rules.
MOVE 02 · SYNTHESIS
The model writes from what it trusts
The model compresses those sources into one answer, leaning on pages with clear, extractable claims. A 2,000-word page with no quotable sentence contributes nothing. A comparison table with real numbers gets lifted almost verbatim.
MOVE 03 · CITATION
A handful of sources get named
Only a few of the retrieved pages earn a citation. Those citations are the new page one, except there are three of them instead of ten, and the buyer never sees the rest.
The TRACE Score. One number for how visible you are to machines.
We run a fixed set of your real buyer questions across the major engines every month, then score five dimensions of what comes back. No vibes, no screenshot of one lucky prompt.
How engines characterize you when they do mention you. Recommended, listed neutrally, or flagged with caveats. Sentiment inside answers behaves differently than sentiment on review sites.
Whether your pages actually enter the working set the engine pulls before answering. This is the earliest and most common point of failure, and it’s invisible from the outside without testing for it.
Which third-party domains feed answers in your category, and whether you exist on them. Engines lean hard on a small set of sources per topic. We map that set and get you into it.
Your share of voice inside answers versus named competitors, tracked across the same question set month over month. This is the number that moves when the work is working.
Whether models resolve your brand as a distinct entity with correct facts attached. Wrong founding year, dead product names, confusion with a similarly named company. Entity rot is common and fixable.
Every engagement opens with a TRACE baseline. Every month closes with the delta. If the score isn’t moving, we change the plan, not the reporting format.
What we actually do between the reports.
Four workstreams. Each one maps to a failure point in how answers get made.
Prompt-set testing[1]
We build a question set from how your buyers actually phrase things, not how your category page phrases things, then run it across engines on a schedule. Raw answers get archived, so you can watch your presence change answer by answer, not just in a score.
Citation portfolio[2]
Engines cite from a narrow bench of domains per topic. We identify that bench for your category, audit where you’re absent, and run placement work to get you onto the sources the models already trust. Being cited is downstream of being citable.
Quotable structure[3]
We restructure your key pages so they contain claims a model can lift cleanly. Specific numbers, direct comparisons, definitions that stand alone. Prose written for a human skimmer and a machine extractor at once, without reading like it was written for a machine.
Entity repair[4]
We audit how models represent your brand as an entity, then fix the record at the sources models learn from. Structured data, knowledge bases, the third-party pages still repeating your 2019 positioning. Models can’t recommend what they misremember.
Three ways to work with us.
Scope depends on how many products, markets and competitors we’re tracking. Pricing is a fifteen-minute conversation, not a form.
Foundation
For teams who need to know where they stand before committing to anything
- Full TRACE baseline across engines
- Citation portfolio map for your category
- Entity audit with correction plan
- Prioritized fix list your team can execute
Growth
For teams who want the score moving every month without building the muscle in-house
- Everything in Foundation
- Monthly prompt-set testing and TRACE delta
- Quotable-structure rewrites on priority pages
- Source placement into your citation bench
- Competitor share-of-voice tracking
Scale
For multi-product or multi-market brands where AI visibility is a board-level number
- Everything in Growth
- Per-market and per-product TRACE tracking
- Quarterly competitive teardown
- Dedicated strategist
Field notes
What we’re seeing in AI answers across categories. Written for practitioners, not the algorithm.
Questions we get on every call.
Parts overlap. Clean structure and authoritative sources help both. But the failure modes are different: you can rank #1 and never enter an engine’s retrieval set, and you can be cited constantly from pages that rank nowhere. We test against the answers themselves, which SEO tooling doesn’t see.
No, and be suspicious of anyone who does. Engines change retrieval behavior without notice. What we guarantee is measurement you can audit: the same question set, run the same way, every month, with the raw answers archived. You see exactly what moved and what didn’t.
Entity fixes and quotable-structure changes can show up in answers within weeks because several engines retrieve live. Citation portfolio work is slower since it depends on third-party placement. A realistic first checkpoint is the 90-day mark.
ChatGPT, Perplexity, Google’s AI Overviews and AI Mode, Gemini, Claude and Copilot. Coverage weighting depends on where your buyers actually are, which the audit establishes first.
Rarely. Most SEO agencies aren’t testing AI answers at all, so the work is additive. Where the same page serves both goals, we coordinate on the edit rather than fight over it.
Find out what the machines say when you’re not in the room.
The audit takes two weeks. You get your TRACE baseline, the raw answers we tested, and a fix list that’s yours to keep whether you engage us or not.