Five-minute video teardown
We record your shelf in three AI models. One business day. No call required.
Request the video → method
We engineered a system because the channel didn't have one yet.
AEO is not advice. It is four stages — Audit, Instrument, Optimize, Compound — running continuously, instrumented end-to-end. Below is every deliverable that ships across a 12-month engagement. Density is the point.
01 AUDIT
what this stage delivers
Before we run our mouth, we run hundred prompts.
We map your category's actual AI shelf before we propose a single change. Hundred buying prompts across four models — ChatGPT, Claude, Gemini, Perplexity, and Google AI Overviews — taxonomized by buyer intent, product subcategory, and competitive set.
You receive the citation matrix, the competitive gap analysis, and a prioritized fix list within forty-eight hours. Engagement or not, the artifact is yours. We've built this audit ninety-three times across four categories — the methodology is locked, the deliverable is consistent, the insights are not.
AUD-01 100-prompt citation matrix Per-engine, per-SKU, vs. top 3 competitors
AUD-02 Vulnerability shortlist SKUs cited zero times by any model
AUD-03 Momentum shortlist SKUs gaining citation share month-over-month
AUD-04 Crawler permissions audit OAI-SearchBot, GoogleOther, PerplexityBot, ClaudeBot
AUD-05 Structured-data audit JSON-LD product schema gaps, OG, RDFa
AUD-06 Category prompt taxonomy 100 prompts mapped to buyer-intent stages
describe this diagram
A six-by-six grid representing hundred buying prompts scored against four AI engines. The top row, marked with a magenta border, is the client brand's results — sparse citations, mixed between magenta dots (where the brand is cited) and ink dots (where a competitor wins the answer). Rows two through five show competitor citation patterns, denser overall. The bottom row fades into the canvas, signaling that one hundred ninety-four more prompts continue beneath the visible sample.
02 INSTRUMENT
what this stage delivers
If you can't measure the channel, you can't argue for the budget.
AI-referred traffic doesn't carry standard referrer headers. Most attribution stacks see it as direct traffic or attribute it incorrectly to organic search. The result is a channel that doesn't show up in any dashboard your CFO trusts.
We install server-side detection, first-party attribution, and an AEO dashboard layer that surfaces in your existing analytics stack. By week three, the question “what did AI search drive last week” gets answered with a number, not a hand-wave. That number is what unlocks the budget conversation.
INS-01 AI-traffic identification Server-side detection by user-agent + referrer fingerprint
INS-02 Citation tracking pipeline Daily prompt sweep, deltas surfaced to Slack
INS-03 AEO dashboard Looker Studio or PostHog — your stack
INS-04 Server-side attribution First-party events to GA4 + warehouse
INS-05 Share-of-voice benchmarks Weekly snapshot, four engines
INS-06 CFO-ready definitions doc What we count, why, with a sample monthly report
describe this diagram
A vertical four-stage pipeline. Top to bottom: chat clients (ChatGPT, Perplexity, Gemini); server-side detection by user-agent and referrer; attribution layer joining session to first-party events; and finally the AEO dashboard — bordered in magenta as the conversion endpoint. Each downward connector is labeled with the operation it performs: fingerprint, attribution, aggregation. The third connector, into the dashboard, also renders in magenta.
03 OPTIMIZE
what this stage delivers
Six surfaces ship in the first sixty days. None of them are blog posts.
Optimization in AEO is engineering work, not marketing work. JSON-LD product schema rebuild. Product feed reconstruction for AI shopping agents. Agentic Commerce Protocol enablement. Universal Commerce Protocol registration. Crawler permissions audit. Structured FAQ deployment.
Every shipped surface is scheduled with your engineering team, code-reviewed against your engineering team's standards, and measured against citation outcomes — not vanity content metrics. We don't ship anything we wouldn't want our own name attached to in a postmortem.
OPT-01 JSON-LD product schema rebuild Spec: schema.org/Product, latest validators
OPT-02 Product feed for AI agents ACP-2025-09 compliant, Merchant-Center-mirrored
OPT-03 ChatGPT Instant Checkout Agentic Commerce Protocol enablement, Stripe integration
OPT-04 Google AI Mode / Gemini Universal Commerce Protocol registration
OPT-05 Perplexity Merchant Program Submission, asset hand-off, citation review
OPT-06 Citation seeding Targeted placement on the third-party platforms LLMs pull from
OPT-07 PDP rewrite for AEO Question-answer structure, comparative tables, structured FAQs
OPT-08 Review schema Aggregate + individual review structured data on every PDP
OPT-09 Crawler permissions ship robots.txt + meta directives, per engine
OPT-10 Sitemap for AI Daily-refresh sitemap, prioritized by margin
describe this diagram
A product detail page wireframe at the center: image placeholder, title bar, price bar, and three stacked description blocks. Six numbered callouts radiate from the wireframe — three from the left, three from the right — each connected by a hairline leader line terminating in a magenta dot on a specific PDP surface. The callouts label, in order: JSON-LD product schema, structured FAQ block, review aggregation markup, server-rendered description, comparative attribute table, crawler permissions header.
04 COMPOUND
what this stage delivers
Share-of-voice is the only metric that survives the next eighteen months.
Citations compound. Third-party mentions stack on top of each other. Models retrain on the previous quarter's open web. The brands that started this work in early 2026 will have a structural lead by mid-2027 that late entrants cannot close inside their planning horizon.
Compound-phase work is monthly: citation seeding on the third-party platforms LLMs draw from, schema regression checks against new model releases, quarterly business reviews tied to citation share, annual matrix re-runs on the anniversary. The work is unglamorous. The trajectory is not.
COM-01 Monthly citation report Share, deltas, competitive shifts, prompts won/lost
COM-02 Quarterly business review Revenue attribution, margin analysis, forward roadmap
COM-03 Editorial cadence 1 long-form research piece / month, citation-seeded
COM-04 Schema regression watch Daily structured-data validity check across catalog
COM-05 New-engine readiness First mover on every commerce-protocol spec that ships
COM-06 Annual audit refresh Full 100-prompt matrix re-run on the anniversary
describe this diagram
A twelve-month line chart of citation share. The competitor average starts near eighteen percent in month one, plateaus around twenty-two percent through month five, then descends to nineteen percent by month twelve. The client brand line, in magenta, starts at six percent, accelerates through months three through seven in an S-curve, and finishes near thirty-four percent. The two lines cross around month five — marked with a magenta dot and the label “crossover · month 5”.
the work we turn down
Trust is built by closing doors.
- no
We don't write blog posts you'll be embarrassed to publish.
We've watched too many agencies bury their clients in 1,200-word “thought leadership” no one cites and no one reads. That work doesn't move citation share, and we don't pretend it does.
- no
We don't promise rankings inside an LLM.
There is no ranking inside a language model. There are citation probabilities, and they shift weekly. Anyone selling “rank #1 on ChatGPT” is selling a fiction we won't co-sign.
- no
We don't run keyword-stuffed PDP copy in the name of “AEO.”
The same tactics that killed early-2010s SEO will kill early-2026 AEO. Your product page is read by humans first and models second. We optimize for both — in that order.
- no
We don't onboard a brand we can't measurably move inside ninety days.
If your category is already locked, your catalog is too thin, or your engineering team can't ship structured-data changes inside a quarter, the engagement will fail. We'd rather tell you on the audit call than nine months in.