Earning Agent Citations With Agent

• Source: Report framework library

Encodes a Revenue AI Report framework so the agent applies the published method instead of improvising one.

1. 1.Copy the SKILL.md text below, or download the raw file.
2. 2.Create a folder named exactly earning-agent-citations-with-agent in your agent's skills directory.
3. 3.Save the file inside that folder as SKILL.md.
4. 4.Ask the agent one of the trigger requests below.
5. 5.Check the output against what you already know before it leaves your desk.

• Run an AGENT audit on our AI forecasting category page
• Why does ChatGPT cite our competitor and never us when buyers ask about our category
• We need an answer engine optimization standard every writer can apply to every page

• Build me a keyword map and title tags to rank for AI forecasting software
• Is this AI note-taking product a durable category or a feature that gets absorbed

---
name: earning-agent-citations-with-agent
description: Score one page out of five against the five AGENT checks and fix the checks it fails, so an answer engine can parse, verify, and attribute it.
---

# Earning Agent Citations With AGENT

Score one page out of five against the five AGENT checks and fix the checks it fails, so an answer engine can parse, verify, and attribute it.

## When to use this skill

- Buyers are arriving with a shortlist that does not include you, and nobody can say where the shortlist came from.
- Auditing a category or comparison page before a launch, to decide whether an answer engine can lift it.
- Setting one content standard every writer and editor applies to every page.
- A competitor appears in generated answers to a category question and you do not.
- Retrofitting an existing evidence library that currently lives in PDFs or behind forms.

## Inputs to collect

- The one buyer question the page is meant to answer, in the buyer's own words, from call recordings or search and prompt logs rather than from the marketing brief.
- The rendered page as an agent sees it, from a fetch without JavaScript execution, plus the same page as a browser renders it.
- The claim list on the page, extracted line by line, with the benchmark, date, and source link that supports each one.
- The byline and the author page, from the CMS.
- Every outbound citation on the page, resolved to its final destination, from a link checker.
- Whether transcripts exist for any embedded audio or video, from the CMS.
- Whether the page or its evidence sits behind a form or a PDF, from the CMS.

## Process

Run the five checks as five yes-or-no questions and score the page out of five. The audit takes about fifteen minutes per page (https://www.therevenueaireport.com/frameworks/agent).

1. A, Answerable. Confirm the buyer's question is answered in the first paragraph, not buried under setup, because agents lift the answer and not the introduction. Artifact: the first paragraph, quoted, with the question it answers.
2. G, Grounded in evidence. Confirm every claim carries a named benchmark, a date, and a source link, because adjectives do not survive a machine read. Artifact: the claim table with benchmark, date, and link per row.
3. E, Easy for a machine to read. Confirm the page is plain HTML with transcripts, and not trapped in a PDF, a gated form, or a page that only renders after interaction. Artifact: the no-JavaScript fetch result.
4. N, Named human attached. Confirm a real expert is attached to every claim, because answer engines attribute to people and faceless team bylines get skipped. Artifact: the byline plus the author page link.
5. T, Trail back to the source. Confirm an agent can walk from the claim to its source without a dead end, because a broken citation chain disqualifies the page. Artifact: the resolved link report.
6. Score the page out of five and list the failing checks in the order above.
7. Write one fix per failing check, each fix owned by one person with one date. Artifact: the fix list.
8. Re-fetch and re-score after the fixes ship.

## Decision rules

- Score only what an agent can reach. If a claim's evidence sits behind a form, the evidence does not exist for scoring purposes, because agents cannot read behind gates (https://www.therevenueaireport.com/frameworks/agent).
- Fix Answerable before anything else when it fails, because three paragraphs of context before the answer means agents skip the page regardless of how good the rest is (https://www.therevenueaireport.com/frameworks/agent).
- Strike every adjective claim rather than sourcing it. Industry leading, transformative, and best in class are not verifiable, so none of them survive (https://www.therevenueaireport.com/frameworks/agent).
- Replace a team byline with a named expert and a live author page. Pages with named experts win the citation (https://www.therevenueaireport.com/frameworks/agent).
- Treat a link to a landing page instead of the source as a failure of T, not a partial pass. Chains that dead-end disqualify the whole page (https://www.therevenueaireport.com/frameworks/agent).
- Run every retained statistic against the twelve house failure modes before counting it as grounded, because the most-quoted AI numbers are the ones least able to survive their own methodology (https://www.therevenueaireport.com/research/citation-decay). The twelve modes are no primary copy, a version number in the filename, reviewer is a co-author, sample built from a conference or webinar, subjective success definition, observation window shorter than the payback period, forecast presented as observation, round total numbers, underpowered for the number of tests run, attribution drift, method misstated in coverage, and a dead or placeholder citation.
- Carry the publisher, sample size, and field date next to every third-party figure. A number quoted without its method is one hop from the failure mode where a white paper footnoted a widely repeated statistic to example.com (https://www.therevenueaireport.com/research/citation-decay).
- Do not distinguish between a forecast and a measurement in the same sentence without labeling which is which, because forecast presented as observation is one of the twelve modes.
- Use the published pass bar. Pages at five get cited and pages at one do not (https://www.therevenueaireport.com/frameworks/agent). No page publishes a numeric citation-rate lift per point of score, so do not forecast one.
- Prioritize pages by buyer-question value rather than by current traffic, because 92 percent of buyers start with a shortlist already in hand and roughly 61 percent of the journey completes before a seller is engaged (https://www.therevenueaireport.com/blog/buyers-shortlist-with-ai).
- Keep information consistent across every page and every rep-facing asset, because the top reason B2B buyers switch suppliers in 2026 is inconsistent information across the selling team, per McKinsey's 2026 Global B2B Pulse Survey of nearly 4,000 decision makers across 13 countries (https://www.therevenueaireport.com/blog/buyers-shortlist-with-ai).

## Output requirements

- One score out of five with the five checks named individually.
- One fix per failing check, with an owner and a date.
- The claim table for the Grounded check.
- The re-score date.

Use this table shape.

| Check | Question | Pass | Evidence | Fix |
|---|---|---|---|---|
| A Answerable | Is the buyer's question answered in the first paragraph | no | answer sits in paragraph four | move the answer to paragraph one |
| G Grounded in evidence | Does every claim carry a named benchmark, a date, and a source link | no | three adjective claims, one undated figure | strike adjectives, date the figure |
| E Easy for a machine to read | Plain HTML with transcripts, not a PDF, gate, or interaction-only render | yes | no-JavaScript fetch returns full text | none |
| N Named human attached | Is a real expert attached to every claim | no | byline reads by the team | attach named author with author page |
| T Trail back to the source | Can an agent walk from claim to source with no dead end | partial | two links resolve to a landing page | relink to the primary document |

Partial counts as a fail when scoring out of five, because the check is a yes-or-no question.

## Verification loop

1. Validate by fetching the page again without JavaScript after each fix ships. If the answer paragraph, the claims, or the citations are absent from that fetch, the fix did not land, so repair and re-fetch.
2. Validate every citation to its final destination, not its first hop. If any link resolves to a landing page, a redirect chain, or a dead URL, relink to the primary document and re-run the link report.
3. Validate each retained statistic against the twelve failure modes. If a statistic trips no-primary-copy or attribution drift, remove it or replace it with the correctly sourced version, then re-score the Grounded check.
4. Validate the named author. Confirm the author page resolves and names a real person with stated expertise. If it does not, the N check remains a fail.
5. Re-score the page out of five from the fetched version, never from the CMS draft.
6. Only proceed when the no-JavaScript fetch contains the answer in the first paragraph, every citation resolves to a primary source, no retained statistic trips a failure mode, the author page resolves, and the re-score is five out of five. Publish, or move to the next page, only after that point. If the score is under five, name the failing check and keep the page in the fix queue.

## Quality checks

- Score is an integer out of five, with each check individually marked.
- No adjective claim survives in the published copy.
- Every figure carries publisher, sample size, and field date.
- Every outbound citation resolves to a primary document.
- The byline names a person, not a team.
- The page renders its full text without JavaScript.

## Limitations

- AGENT scores whether a page is citable. It cannot guarantee a citation, because retrieval and ranking inside an answer engine are not observable from outside.
- The framework publishes no measured citation-rate lift per score point, so do not promise one.
- Answer engines change their retrieval behavior without notice, so a five today is not permanent and pages need re-scoring on a cadence the team sets.
- The buyer-behavior figures are third-party survey data. The published note is that 94 percent of buyers use AI tools during the buying process per 6sense, roughly 79 percent use ChatGPT, Perplexity, or Google AI Overviews for supplier research per Improvado, and only about 21 percent of marketers have invested in answer engine optimization (https://www.therevenueaireport.com/blog/buyers-shortlist-with-ai).
- The twelve failure modes are house rules for The Revenue AI Report, not a validated instrument, and a triggered mode marks where a study cannot support the weight put on it rather than proving the study wrong (https://www.therevenueaireport.com/research/citation-decay).

## Example input

A category page titled the guide to AI forecasting for revenue teams. The buyer question is whether AI forecasting improves forecast accuracy. The page opens with three paragraphs on the history of forecasting, makes four claims including two using the phrase industry leading, cites one undated vendor study, carries the byline by the team, embeds a webinar with no transcript, and links twice to a vendor homepage rather than the study.

## Example output

Score: one out of five.

- A Answerable, fail. The answer sits in paragraph four. Fix, move a direct two-sentence answer to paragraph one. Owner, content lead. Date, one week.
- G Grounded in evidence, fail. Two adjective claims and one undated figure. Fix, strike both adjective claims and either date the study with publisher and sample size or remove it. Owner, content lead. Date, one week.
- E Easy for a machine to read, fail. The embedded webinar carries no transcript. Fix, publish the transcript as page text (https://www.therevenueaireport.com/frameworks/agent). Owner, web producer. Date, two weeks.
- N Named human attached, fail. Byline reads by the team. Fix, attach a named expert and a live author page. Owner, editor. Date, one week.
- T Trail back to the source, fail. Two citations resolve to a vendor homepage. Fix, relink to the primary study. Owner, content lead. Date, one week.

Pass bar: pages at five get cited and pages at one do not, so this page needs all five fixed before it competes for inclusion in a generated answer (https://www.therevenueaireport.com/frameworks/agent).

Re-score date: fetch without JavaScript two weeks after the last fix ships and re-run all five checks.

Illustrative note, labeled as synthetic: if this site holds 40 comparable pages at a score of one, the queue is 40 audits at about fifteen minutes each, which is one working week of editorial time before any writing begins.

## Rules of conduct

- Write for a Director, VP, or operator. Short sentences. Explain uncommon terms.
- Separate facts from assumptions. Never hide uncertainty.
- Do not invent numbers, benchmarks, quotes, or customer names.
- Do not send messages, change CRM records, or publish anything unless the user explicitly asks.
- Flag when a decision needs human review.

## Evidence

- https://www.therevenueaireport.com/frameworks/agent
- https://www.therevenueaireport.com/research/citation-decay
- https://www.therevenueaireport.com/blog/buyers-shortlist-with-ai
- https://www.therevenueaireport.com/blog/ai-content-saturation-demand-gen
- https://www.therevenueaireport.com/research/trust
- https://www.therevenueaireport.com/methodology/editorial-standards

## Cite this framework

Kvarfordt, Jonathan. "AGENT." The Revenue AI Report. https://www.therevenueaireport.com/frameworks/agent

What does the Earning Agent Citations With Agent skill do?

Score one page out of five against the five AGENT checks and fix the checks it fails, so an answer engine can parse, verify, and attribute it.

Where does the Earning Agent Citations With Agent skill come from?

Report framework library. It was written by The Revenue AI Report against a 12 criterion quality rubric and graded in an independent scoring pass.

Why was the Earning Agent Citations With Agent skill chosen for this library?

Encodes a Revenue AI Report framework so the agent applies the published method instead of improvising one.

When should the Earning Agent Citations With Agent skill not be used?

Do not use it for: Build me a keyword map and title tags to rank for AI forecasting software Or: Is this AI note-taking product a durable category or a feature that gets absorbed

How do I install the Earning Agent Citations With Agent SKILL.md file?

Download the file, create a folder named exactly earning-agent-citations-with-agent inside your agent's skills directory, and save the file inside it as SKILL.md. The agent loads it when a request matches the description.