Live AI Citation Rates for aeoforged.com (Updated Daily)

Ryan KingsFounder & CTO

Published August 2026 · 14 min read

This page publishes what our visibility instruments recorded for aeoforged.com — branded vs generic citation rates from daily real-engine checks — then the programme philosophy (four client-facing layers) and the system layers (how platform tools, orchestrating agents, and humans split). Live aggregates sit under the title and refresh as checks land. Independent retests are welcome; the full observation pack is available on request after review.

What are the limitations of this data?

Treat every number on this page as first-party dogfood, not an independent lab study.

These figures come from a cold start: solo founder, no external funding, no pre-existing brand authority or team. The domain and content were built from zero in 2026. That context is why the numbers stay labelled first-party dogfood and directional — not because the instruments are soft, but because the starting conditions were.

1. Short measured window. The live board’s check history for this workspace begins in early August 2026 (see the snapshot asOf / last-check stamps). It is not a multi-month authority study.
2. First-party only. We built the content and the instruments, then measured ourselves. Useful feedback; not outsider validation.
3. Small unbranded absolute count. Generic page-tier query wins are few (see the live “generic page-tier wins” count). Absolute client voices behind generic SOV are also small when the snapshot shows them that way.
4. Engine concentration. On this board, generic checks that reached page tier have been Perplexity-only so far (other engines: zero page-tier hits on those generic queries in the same sample). Cross-engine consistency is the harder bar.
5. Query selection bias. Tracked prompts are operator-chosen for this programme. Wins can cluster on narrow technical wording; broader category prompts may still be gaps — read the live board, do not assume a category census.
6. Coverage ≠ citation. Answer coverage means an extractable page candidate exists in our join. It does not mean engines cite that page.

What do the instruments show?

Live aggregates sit above this section. Narrative below is framing for those instruments — when rates move, trust the snapshot, not frozen prose.

Branded standing is plumbing, not category authority. Branded prompts contain a brand term; the live branded citation rate is on the snapshot. A high branded rate shows entity/primary-source plumbing can work. It is not proof of generic category authority.

Generic / goal questions: thin page-tier wins. The snapshot’s generic page-tier win count is the measured absolute (query-status wins at page tier). A small non-zero count is real; it is not “half the category board.” Whether broader prompts (“best AEO tools”, “how to get cited in AI answers”) are still gaps is a live board fact — check the widget and, if needed, the pack’s query list.

Share of voice on generic queries is directional when voices are small. The snapshot reports generic-basis SOV % plus client vs competitor voice counts (not a named competitor leaderboard). We deliberately omit a competitor rank table on this public page.

What the measurement layer does in code. Checks store client_tier (page | domain | brand | none). The scoreboard splits branded vs generic slices and, when a class map exists, computes headline SOV on generic checks only. Completed checks that miss still persist (none). Failed engine calls are not written as miss rows — a day with no row means no completed check that day.

For category-level context beyond our domain, see ChatGPT often answers without sources, AI sector reports, the sector citation benchmark explainer, and the four-study citation-economy synthesis. On-site score lift is a separate claim in the 8→81 case study. A separate first-party note on the llms.txt generator — referrers, Search Console, and the citation ledger — is ChatGPT referral traffic on an llms.txt generator.

How can researchers retest or get the data pack?

Public on this site: aggregate rates and counts in the live snapshot, this methodology prose, and published studies under /research.

On request (walled pack): after operator approval, a private download link (90-day aeo_dp_ token) for an auto-generated zip: queries.csv, observation_ledger.csv (per completed check; no answer excerpts), aggregates_daily.csv / aggregates_weekly.csv, protocol.md, scoring_rubric.json (score version + dimension weights by page type), and README.md. Pack daily/weekly “cited” counts treat page|domain|brand as cited — that can differ slightly from page-only board green.

Request the pack with a work email and a one-line research purpose. We log requests, apply a light review (research / independent analyst), and aim to respond within 48 hours.

What is the four-layer programme?

The numbers above come from Layer 03 (Monitoring). They only make sense inside a sequence we run the same way for dogfood and for clients. This is the client-facing programme — not the same taxonomy as the system layers in the next section.

Why sequence matters. Content without a readable site is noise. Authority without measurement is storytelling. Monitoring without a next page to write is a dashboard. Distribution without a grounded canonical URL is busywork. The layers are ordered so each one supplies the precondition for the next — and so we never sell “N fixes” or a citation guarantee as the product unit.

LAYER 01 · AI readiness

Philosophy: if machines cannot fetch, parse, and extract who you are and what you offer, later work does not stick.

In product terms: complete-audit standing (content AEO + technical reachability), the AI-Ready capability checklist (access, machine meaning, answer-ready pages, trust signals — deterministic ratings, not model grades), and agent-operability probes (what agent user-agents can do on the site today). Itemized fix lists unlock through Fix Programme / quoted we-apply work; verify-page re-checks live URLs against a pinned baseline. Who applies: client agent, human/dev team, or negotiated we-apply — we measure.

LAYER 02 · Authority

Philosophy: engines need a clear picture of your brand — not a lookalike or an off-category association. Entity clarity is a different problem from “write more posts.”

In product terms: Authority OS — profile categories (entity identity, people, topical depth, primary evidence, distribution, off-site corroboration, AI recognition), measured P0–P2 phase checks where the engines are deterministic, plus plan-only external work that is never pretended to be measured. Binding constraint language names the next failing category. Authority compounds slowly; we report that rather than inventing a DA-style lift.

LAYER 03 · Monitoring

Philosophy: green only when a page of yours is cited for a tracked question. Brand mentions and domain-only hits are different tiers. Misses are results.

In product terms: the citation board (branded lane + goal/generic lane), daily real-engine checks, branded/generic rates kept separate in claims, generic-basis share of voice when a class map exists, strategy reports composed from those measured inputs. The live snapshot on this page is Layer 03 applied to aeoforged.com. A monitoring-only month can honestly show flat generic lines — that is allowed copy, not a failure of the instrument.

LAYER 04 · Content & distribution

Philosophy: write for gaps the board (or strategy) already named; ground the draft in retrieved research; score for publish-readiness; ship a canonical URL; package socials for clients to post. We do not take social logins and we do not promise reach.

In product terms (Grow/Dominate): gap-tied content picks → platform web research (citations limited to retrieved URLs) → client agent or human writes → score / improve loop → publish live URL → distribution kit (LinkedIn / X / newsletter checklists). House split: we research and score; your agent writes (unless a human editor owns the draft). We package; you post. Scores are feedback, not a promise engines will cite the page.

How this page’s dogfood fits. Layer 03 is what you are reading in the widget. Layer 04 is how programme content (including this article’s publish path: repo → PR → deploy → register URL) is supposed to respond when the board shows gaps. Layers 01–02 are the foundation and entity work that make page-tier citations possible at all — they are not proven by a high branded rate alone.

What are the four system layers?

These are machinery and custody — orthogonal to the programme sequence above. Two different “agents” show up — do not collapse them:

1. Platform tools — AEOForged research, score, audit, verify-page, and related runners. They execute when called (REST, MCP, dashboard, or in-app agent loop). They are not the buyer’s IDE agent.
2. Orchestrating agents — the buyer’s own agent (e.g. Cursor, Claude Code via handoff/API key) or our optional session in-app agent. They call tools, draft/edit, and apply under trust scopes. They do not grant verified by checkbox.

On Fix Programme and connected work, verify-page is the only product writer that can set action-item status to verified (PATCH/MCP transitions reject verified with 422). Craft-over-score is binding agent policy (skills / task packs) — not a separate HTTP validator that rejects FAQ chrome. Stacking FAQ mirrors or question-stuffed headings only to move a dimension is still a failed fix even if a score rises.

Related: AEO scoring dimensions, what is an AEO audit, what AI engines look for.

How do agent workflows include humans?

An orchestrating agent bootstraps a scoped handoff (or uses an API key), can load the next task pack from bootstrap, calls platform tools (research, score, …), drafts or edits on the allowed surface (repo, GitHub PR, or CMS), then can call verify-page on a live URL. Humans may edit mid-loop. Durable HITL includes aeo_ask_human (/api/v1/decisions) and article-review share links (aeo_rv_). Defaults in code: propose_changes: true, ship_live: false until a human changes trust on the room permissions route. Payment and brand-profile Confirm are room/human surfaces — not handoff tool verbs (possession of a room link is still a credential; the gate is not a separate “is human” detector).

Writing is not always “the agent”: a human editor can own the draft. Platform hosted write tools also exist when callers choose them. Shipping live CMS apply requires ship_live; GitHub apply is propose/PR under propose_changes.

A Dominate-shaped content month as productized: content picks (room/API) → agent or human runs research/score/draft via tools → content-lifecycle publish registers a live URL and can generate a distribution kit for clients to post. Internal link mesh and article Approve-via-review are required by task-pack / dogfood guidance (and CI for our own articles); they are not hard blockers inside every publish API call. This paper’s dogfood path was repo → PR → deploy, then register the live URL — not a CMS ship_live apply. See AEO tools for AI agents and AEO reports for AI agents.

Summary

• Lead with measured outcomes and limits; programme philosophy and system custody follow.
• Programme sequence: AI readiness → authority → monitoring → content & distribution (each supplies the next).
• Trust the live snapshot for rates; prose here is methodology and scope, not a frozen scoreboard.
• Pack delivery is approve → 90-day download token → on-demand zip.
• Platform tools measure and verify; orchestrating agents call tools and write under trust scopes; humans keep payment, Confirm, and ship_live.