9 ChatGPT Prompts for AEO Audits and Citation Research

AEO adds a second visibility layer alongside SEO rankings. These nine ChatGPT prompts help you audit brand mentions, compare citation share, map competitor sources, improve extractable passages, and identify question chains. Run them as a repeatable panel across engines and dates, then use the outputs to decide what to measure, publish, update, or promote.

What makes an AEO prompt reliable?

A useful AEO prompt is a reusable research instrument. You can run it with the same variables each month and compare the output without changing the method. A vague request such as “audit my AI visibility” produces an answer you cannot reproduce or measure.

Four design choices make the difference:

• Specific context. Name the category, buyer, use case, brand, and competitors the engine should evaluate.
• Engine coverage. Run the same prompt separately in ChatGPT, Perplexity, and Gemini. Their source pools differ substantially. One analysis found that the engines shared only 14% of their top mentioned sources.
• Stable variables. Use placeholders such as [BRAND], [CATEGORY], and [COMPETITOR] so the wording stays fixed between runs.
• Structured output. Require fixed columns and labels. Consistent output makes changes visible without turning every run into a new interpretation exercise.

This article is the copyable prompt library. Use the deeper AEO audit workflow when you need the technical checklist behind the prompts.

Which AEO prompt should you run for each task?

Each template maps to one decision. Start with the first three to establish a baseline, then use the remaining prompts to turn visibility gaps into content and distribution work.

How should you audit your current AI visibility?

Start with a fixed panel, not a memorable screenshot. Repeated runs matter because generative answers vary even when the wording stays the same. Research on brand recommendations found less than a one-in-100 chance of receiving the same brand list across two runs in the tested systems.

Run the baseline and share-of-voice prompts in clean sessions, record the engine and date, and aggregate the results. The companion guide to tracking ChatGPT citations explains how to turn repeated runs into a stable measurement panel.

Brand visibility baseline prompt

Use this template to record whether the brand appears, its position, the language used to describe it, and the sources attached to each recommendation.

You are simulating a buyer researching [CATEGORY] solutions. Use current web search and complete both parts.

Part 1
1. What are the best [CATEGORY] tools for [ICP] and [USE CASE]?
2. How does [BRAND] compare with [COMPETITOR 1] and [COMPETITOR 2]?
3. Would you recommend [BRAND] for [USE CASE]? Explain the decision criteria.

Part 2
Audit your answers. Record whether [BRAND] appeared, its position, the exact description used, and every cited source that supported each brand.

Return a table with these columns: query, brand, position, description, cited source, source date. State when no source was available.

Keep the wording and variables stable. If you change the category, buyer, or use case between runs, you have changed the test rather than measured a trend.

Share-of-voice check prompt

Share of voice is your brand mentions divided by all brand mentions across a fixed query set. Treat it as directional because the result changes by engine, prompt, and run.

Run each query separately using current web search. List every brand mentioned and preserve its order of appearance.

Queries
1. [QUERY 1]
2. [QUERY 2]
3. [QUERY 3]
4. [QUERY 4]
5. [QUERY 5]

Return two tables. The first should contain query, brand, position, description, and cited sources. The second should contain brand, number of appearances, share of all brand mentions, average position, and common description. Include [BRAND] even when it receives zero mentions.

The description column belongs beside the count. A mention that frames your product as a budget option and one that presents it as the category standard contribute equally to share of voice but imply different work.

How do you reverse-engineer competitor citations?

Citation research should identify the sources an engine trusts and the brands those sources support. Brand web mentions showed a strong correlation with AI visibility across 75,000 brands, which means an authority gap may sit off your website.

B2B citation studies also show that engines regularly use community and review sources. One analysis found Reddit and LinkedIn prominent in B2B answers, while another found a relationship between review-platform presence and ChatGPT citations. These are correlations, not guarantees, but they tell you where to investigate.

Competitor citation mapping prompt

Run this in a session with web search enabled. Repeat it in Perplexity and compare the domains instead of assuming one engine represents the market.

Search the web and answer [CATEGORY QUERY] for [ICP] and [USE CASE]. Then audit every source you cited or consulted.

For each source, record the domain, page URL, source type, publication or update date, brands supported, and the claim it supported. Do not infer hidden sources.

Compare the source list with the visible presence of [BRAND], [COMPETITOR 1], and [COMPETITOR 2]. Identify domains that support a competitor but omit [BRAND].

Return a table with these columns: domain, source URL, source type, supported brand, supported claim, date, gap for [BRAND], possible action. Separate verified findings from hypotheses.

Use the gap list to prioritize actions you can control: complete review profiles, earn inclusion in relevant comparisons, participate in communities, or pitch a source with a specific editorial reason. The guide to how LLMs cite sources covers the retrieval mechanics behind this exercise.

How do you improve content for AI extraction?

Passage structure influences whether a section can stand alone in a synthesized answer. Across 15.7 million Google AI Mode citations, the median extracted passage was 117 words and most extracts put the answer in the first sentence.

The goal is not to force every paragraph into an arbitrary length. Put the answer first, keep the supporting detail self-contained, and preserve the evidence that makes the claim credible. Use the full content optimization workflow when a page needs more than a passage rewrite.

Direct-answer rewrite prompt

Use this on a section that delays its answer, opens with a hedge, or depends on context several paragraphs away.

Rewrite the section below as a self-contained answer block.

Requirements
- Answer the heading in the first sentence.
- Keep the opening answer between 40 and 60 words.
- Preserve every supported fact and its citation.
- Replace vague hedges with details already present in the source.
- Do not invent statistics, examples, or certainty.
- Apply the attached voice guidelines and banned-terms list.

Return the revised block, followed by a short list of facts or citations you could not preserve.

[PASTE HEADING AND SECTION]

Review the output against the source section. The model should restructure supported facts, not manufacture the specificity the original page lacks.

FAQ and schema review prompt

Visible question-and-answer content can improve retrieval when it answers real buyer questions. A reversibility test found that citations rose when FAQs were added and fell when they were removed. Schema is implementation hygiene rather than a guaranteed citation lever. A separate test of 1,885 pages found no citation lift from adding JSON-LD alone.

Review the article below and identify five buyer questions it can answer without adding new claims.

For each question, write a 40 to 60 word answer that leads with the answer and stays within the evidence in the article. Flag any question the article cannot support.

Then generate valid FAQPage JSON-LD that matches the approved question and answer text exactly. If the article contains a genuine ordered process, generate matching HowTo JSON-LD. Do not create steps the article does not contain.

[PASTE ARTICLE]

Only publish questions the page can answer. Put structured FAQ data in the CMS field that generates matching markup rather than duplicating an FAQ section in the article body.

How do you map conversational question chains?

Buyers ask follow-up questions, and engines may fan one query into several retrieval tasks. Pages ranking for related fan-out queries were 161% more likely to receive citations in one study. That makes the missing turns in a buyer conversation useful planning inputs.

Question-chain mapping prompt

Map how [ICP] researches [CATEGORY] in an AI assistant across four stages: problem awareness, solution comparison, decision, and implementation.

For each stage, write one realistic opening prompt and three follow-up prompts that could occur in the same conversation. Map each turn to the best content format and an existing [BRAND] URL. Leave the URL blank when no page answers the question.

Return a table with these columns: stage, opening prompt, follow-up prompt, decision behind the question, best format, existing URL, evidence needed.

Blank URL cells are content gaps, but they are not automatically new-article opportunities. First decide whether an existing page should answer the question. The AI search content strategy guide shows how to separate updates from net-new pages.

How do you build a reusable AEO prompt library?

Store one row per prompt and engine. Record the prompt text, variable values, buyer stage, run cadence, last run date, and output URL. This creates a change log instead of a folder of screenshots.

Validate prompts manually before loading them into a paid tracker. Remove queries that do not resemble buyer behavior, merge duplicates, and keep wording stable for the panel that remains. Any individual prompt is replaceable. The controlled panel and its history are the durable assets.

Which advanced AEO prompts belong in the library?

Add these templates after the baseline panel is stable. They turn recurring patterns into specific research questions without expanding this page into another AEO strategy guide.

Topical authority gap prompt

Using current web search, compare the retrievable topical coverage of [BRAND DOMAIN] and [COMPETITOR DOMAIN] for [CATEGORY].

List the subtopics each domain covers with a specific supporting URL. Do not infer coverage from brand familiarity. Mark a subtopic as unverified when you cannot find a supporting page.

Return a table with these columns: subtopic, [BRAND] URL, [COMPETITOR] URL, coverage gap, buyer stage, recommended update or new page. Prioritize gaps that appear in the question-chain map.

Treat model confidence as a hypothesis. Verify every claimed gap against your site, the competitor site, and current search results before creating work.

Off-site citation check prompt

Using current web search, identify third-party review platforms, communities, publishers, and comparison pages that discuss [CATEGORY]. Record whether each source mentions [BRAND], [COMPETITOR 1], or [COMPETITOR 2].

Return a table with these columns: source, source type, relevant URL, brands mentioned, claim or recommendation, publication date, gap for [BRAND], practical next action.

Do not treat a platform homepage as evidence. Include the specific page where the brand or category appears.

The output should produce an outreach and profile-maintenance list, not a mandate to appear everywhere. Prioritize sources that repeatedly influence your tracked queries and are relevant to your buyers.

Page-role classification prompt

Review these [BRAND] pages for [CATEGORY]: [LIST OF URLS].

Classify each page as a direct-answer target, which should resolve a focused question in a self-contained passage, or a synthesis target, which should contribute evidence and perspective to a broader answer.

Return a table with these columns: page, classification, target question, strongest existing passage, missing evidence, structural problem, recommended fix. Support every judgment with text from the page. Do not recommend a new page when an update can close the gap.

Use the classification as a planning lens. A direct-answer page still needs evidence, and a synthesis page still needs clear passages. The distinction only tells you where each page should concentrate its effort.

If you want one practical AI search and content-operations lesson each week, subscribe to The Messy Middle newsletter. It gives you tested workflows and research updates without adding another dashboard to maintain.