Answer Engine Optimization, Step by Step: Turning a Measured Citation Gap Into Owned Keyword Work

Answer Engine Optimization, Step by Step: Turning a Measured Citation Gap Into Owned Keyword Work

What we found on 2026-08-24: the TrustGrowth retrieved census recorded 0 citations (empty citations list) for this site. The recalled sample on the same date published a mention rate of 0.2393 with Wilson interval 0.119–0.423, n = 27.7893, answered_count 30. This tutorial shows how that dated citation gap becomes a tracked keyword target and then a page-level brief. It does not claim any later lift.

Prerequisites

Before starting this walkthrough, confirm you have the following:

• Access to a TrustGrowth account connected to Google Search Console for the site under review. Google documents those Search Console metrics at the Search Console performance report.
• Basic navigation familiarity with the AI Visibility panel, specifically its retrieved census, recalled sample, readable score, reach, and impact views.
• An understanding of Search Console page and query data (impressions, clicks) as surfaced through TrustGrowth's impact pull.
• No coding required. The JSON snippets below are illustrative field shapes. They are not measured records, and they are not code you execute.
• General familiarity with what E-E-A-T (Experience, Expertise, Authoritativeness, Trustworthiness) means. This piece does not redefine it.

Method and limits

Every statistic in this article is quoted from a named TrustGrowth pull. We do not re-derive n, the Wilson interval, or the mention rate here.

• Retrieved census, 2026-08-24: citation count 0, citations list empty. Method question: has any engine returned this domain as a source, for this tracked query set, as of that pull.
• Recalled sample, 2026-08-24: mention rate 0.2393; Wilson interval 0.119–0.423; n = 27.7893; answered_count 30. Method question: across a fixed run of prompts, what proportion mentioned the domain.
• Recalled sample by engine, 2026-08-24, n = 10 answered per engine: Anthropic 10/5/5, OpenAI 10/5/5, Gemini 10/0/10 (answered / mentions / non_mentions). Prompt-set scope is the recalled-sample run on that date, not a general ranking of those products.
• Readable-score crawl, 2026-08-24: overall 95; structured_data 100; authorship_trust 100; answer_structure 81.66; citation_quality 99.13.
• Reach crawl, 2026-07-26: 10 bots present; llms.txt absent. For the reachability check itself, see how to check whether AI crawlers can access a site.
• Impact pull, 2026-08-24: not_measured (0 impressions, 0 clicks). That is an absence of Search Console data, not a measured zero.

Limits: n = 27.7893 is small, so the interval is wide. At n = 10 per engine, one flipped answer moves the engine rate by 10 percentage points. We did not invent a conversion, ranking, or citation-rate outcome. Public proof for the current TrustGrowth snapshot is at the TrustGrowth proof page.

Introduction

Answer engine optimization (AEO) is the discipline of structuring and prioritising content work so that one specific URL can be identified as the citable source for one specific question inside an AI-generated answer. That is a page-level definition. Three adjacent terms mean different things.

Generative engine optimization (GEO) is the in-page cluster sibling of AEO in this piece: influencing how a generative summary draws on multiple sources at once, not a separate discipline. Traditional SEO (search engine optimization) is the older discipline of ranking a URL inside a results list. "AI visibility" here means only one thing: appearing as a cited source in an AI-generated answer, as measured by a specific retrieval or recall method stated inline every time the term is used.

This tutorial is a workflow walkthrough. It shows how a held citation-gap measurement becomes a tracked keyword target and then a page-level brief inside TrustGrowth's pipeline. It is not a structured-data checklist.

Step 1: Start from a measured citation gap, not a new test

The goal of this step is to fix the current-state facts the rest of the workflow acts on. Each figure is bound to the method that produced it and the date it was captured.

Caption: TrustGrowth AI Visibility pulls used in this tutorial. Dates and methods are part of the claim.

The rule for this table: no number appears without its method sitting in the same line. A retrieved census and a recalled sample answer different questions. Conflating the two is the most common misread of this kind of report. Source for the Search Console impact fields: Search Console performance report.

Step 2: Read the gap for what it does and does not say

Zero citations recorded is not the same claim as zero probability of citation, and neither is the same claim as page failure. The recalled-sample interval of 0.119 to 0.423 around the published point estimate 0.2393, at n = 27.7893, is the uncertainty band around that mention rate. It is wide because the sample is small, not because the method is unreliable. Quote the interval whenever you quote the point estimate.

The engine split of 5 mentions out of 10 answered for one general-purpose conversational assistant, 5 out of 10 for another, and 0 out of 10 for a third is an observed asymmetry, not a ranked comparison. At n = 10 per engine, a single flipped answer moves the rate by 10 percentage points, which is too coarse a base to attribute a cause.

Measured vs. not yet known

Step 3: Convert the gap into tracked keyword targets

A measurement sitting in a dashboard is not yet work. From the AI Visibility panel, "Track all as keyword targets" creates one keyword target row per query behind the citation gap. "Open as briefs" then creates a corresponding content brief for each tracked target. Both actions are gated by fit rules specific to the site's category. Rejected queries stay visible as filtered-out, so you can audit what was excluded and why.

A third state matters here. Below a floor of 3 samples, a query's citation status is shown as "unknown", distinct from "not cited." This stops the 0-citation figure from Step 1 being over-read as failure when the underlying sample is too thin to say anything at all.

Illustrative field shape only, not a measured record:

{
 "keyword": "how to structure e-e-a-t signals for a saas pricing page",
 "source_gap_reference": "retrieved_census_2026-08-24",
 "verdict": "tracked",
 "status": "active"
}

Step 4: Turn a tracked keyword into page-level work using Wave 3 rules

A tracked keyword still leaves three questions open: what to write, on which URL, and why. TrustGrowth's Wave 3 rules from the 2026-08-21 dogfood run resolve those three questions per query, outputting an owning_url, an action (one of revise, consolidate, or new_page), and a rationale.

We recorded this batch context on that dogfood run: planning pool 133, qualified 67, Search Console page×query data contributing 45 rows across 18 distinct queries, and 104 site pages considered as candidate owners. These pool literals are current-state facts drawn from the batch. They are not re-derived here.

Illustrative Wave 3 row for the example query, not a screenshot:

{
 "keyword": "how to structure e-e-a-t signals for a saas pricing page",
 "action": "new_page",
 "owning_url": null,
 "rationale": "No page among the 104 considered returns a direct answer to this query; 45 GSC page x query rows across 18 distinct queries show adjacent coverage but no owning match, so consolidation and revision were not selectable actions."
}

Step 5: Read the rationale field before writing anything

action: new_page with an empty owning_url means nothing in the 104-page set currently answers this query, so the rule set had no candidate to revise or consolidate into. That is a constraint on the rule set's inputs, not a judgement about content quality elsewhere on the site.

A rationale you could paste into a brief ticket:

"This query is being opened as a new page rather than a revision because the Wave 3 batch of 2026-08-21 found no owning URL among the 104 pages considered. The planning pool for that run held 133 candidate queries, of which 67 qualified for evaluation; Search Console contributed 45 page-by-query rows spanning 18 distinct queries as supporting signal. None of those rows mapped this query to an existing page with sufficient topical match, so new_page was the only action the rule set could return. This brief should be written to directly answer the tracked query, not adapted from an existing asset."

Step 6: Build the reusable operator checklist

The value of this workflow is that it repeats. Use this sequence on the next gap, not just this one.

No step in this checklist promises an outcome. Each output is labelled by what it actually is: a measurement, or a plan awaiting the next measurement. For a Search Console-first audit sequence that starts the same way, see the GSC audit playbook.

Common pitfalls

• Treating 0.2393 as a stable score. It is a point estimate at n = 27.7893 with a Wilson interval of 0.119–0.423. Restate the interval every time you cite the rate.
• Reading a 0/10 engine split as a verdict on that engine generally. At n = 10, it is a small, engine-specific sample on one prompt set, one date.
• Skipping the fit-rule gate. Manually adding a keyword target that the category fit rules would have filtered out reintroduces the noise those rules exist to remove.
• Confusing "unknown" with "not cited." Unknown means below the 3-sample floor; not-cited means measured, above threshold, and zero.
• Presenting a Wave 3 rationale as a finished brief. It is a starting justification for why the action was chosen, not a content outline.
• Citing Impact figures as measured when the pull says not_measured. The 2026-08-24 impact pull returned 0 impressions and 0 clicks under a not_measured status.

FAQ

What is answer engine optimization?

Answer engine optimization is page-level work so that one URL can be identified as the citable source for one question inside an AI-generated answer. It is not the same as ranking in a results list.

What is the difference between a retrieved census and a recalled sample?

A retrieved census asks whether any engine returned this domain as a source for a tracked query set. A recalled sample asks what proportion of a fixed prompt run mentioned the domain. The 2026-08-24 census recorded 0 citations; the recalled sample recorded mention rate 0.2393 (interval 0.119–0.423, n = 27.7893).

When is a query unknown?

Below a floor of 3 samples, TrustGrowth shows citation status as unknown. Unknown is not the same as not cited.

What does new_page mean?

In the 2026-08-21 Wave 3 rules, new_page with a null owning_url means no page among the 104 considered answered that query, so revise and consolidate were not selectable.

Summary and next steps

The chain runs in one direction: a measured citation gap becomes a tracked keyword target, gated by category fit rules, which becomes a page-level work item with an owning_url, action, and rationale, which becomes a content brief.

No lift, rate-change, or traffic outcome is claimed. The only valid way to see whether the picture in Step 1 has shifted is the next dated pull of the same AI Visibility panel, using the same method labels shown here.

Last measured: retrieved census and recalled sample 2026-08-24; reach crawl 2026-07-26; Wave 3 dogfood 2026-08-21. Next re-run: the next dated AI Visibility pull using the same method labels.