Short answer: To run a manual AI visibility teardown in 30 minutes, choose 5 to 10 buyer-intent prompts, test them across a few AI systems, record whether your brand appears, note which competitors are mentioned, capture cited sources, check whether your brand is described accurately, and compare the results against your site's answer infrastructure. The goal is to find patterns worth investigating, not to prove causation from a single test.
You do not need a platform to get a first read on AI Search Visibility.
You need a structured 30 minutes. A manual teardown will not give you a stable trend or a causal conclusion, but it will surface prompts, competitors, source gaps, inaccurate descriptions, and missing proof that deserve a deeper audit or a build plan.
This is the hands-on companion to What an AI Search Visibility teardown looks for.
A 30-minute teardown should find patterns, not prove causation
Set expectations before you start.
AI systems vary their answers across sessions, phrasings, and time, so one result is a snapshot, not a verdict. Treat everything you find as directional evidence to investigate, and label your findings honestly using the Public Proof Package. A missing mention is a prompt to dig deeper, not proof that the brand is failing.
The 30-minute breakdown
Keep the timer honest:
- 5 minutes: choose prompts
- 10 minutes: run prompts across systems
- 5 minutes: record competitors and citations
- 5 minutes: inspect your own site for matching answer infrastructure
- 5 minutes: summarize gaps and next steps
Step 1: Choose buyer-intent prompts
Pick 5 to 10 prompts a real buyer would ask.
Cover four categories so you see more than one angle:
Category prompts
- Best companies for [category]
- Who helps with [problem]?
- What tools improve [outcome]?
Comparison prompts
- [Brand] vs [competitor]
- Best alternative to [competitor]
- What is the difference between [category A] and [category B]?
Brand prompts
- What does [brand] do?
- Is [brand] good for [use case]?
- Who is [brand] for?
Problem prompts
- Why is my brand not showing up in AI search?
- How do I improve AI search visibility?
- How do I get cited in AI answers?
Step 2: Test across multiple AI systems
Do not rely on one model.
Run the same prompts across a few systems, such as ChatGPT, Perplexity, and Gemini, plus Google AI Overviews where relevant. Different systems synthesize from different sources, so testing more than one is how you separate a system quirk from a real pattern.
Run the free EntityMesh scan alongside your manual test to compare your notes against a structured diagnostic.
Step 3: Record brand mentions, competitor mentions, and source citations
Write down what you see, not what you remember.
For each prompt, capture: whether your brand appeared, where it ranked in the answer, which competitors were named, and which sources were cited. Save the raw output or a screenshot so you can compare across sessions later. Competitor and citation patterns are often more revealing than your own mention count.
Step 4: Check answer accuracy and definition drift
Appearing is not the same as being described correctly.
Read how each system describes you. Is the category right? Is the positioning current? Are old names or retired descriptions still showing up? This gap between how you define yourself and how AI systems describe you is definition drift, and it is exactly what EchoScan monitors over time.
Step 5: Compare prompt results against your website
Now connect the answer to the source layer.
For each gap you found, ask whether your site even answers that question on a crawlable page. If an AI system got your category wrong, is the correct category stated clearly and consistently on your site? If a buyer question went unanswered, does a public answer for it exist anywhere, or only in sales calls?
Step 6: Look for missing Support Hub and Answer Hub content
Map gaps to missing infrastructure.
Many teardown findings trace back to the same root cause: the best answer is not published on a crawlable page. Note which prompts should map to a Support Hub or Answer Hub page that does not exist yet. Those become build candidates.
Step 7: Label findings as Proof-Grade, Directional, or Not Yet Measured
Grade what you found before you act on it.
- Proof-Grade: the page exists or it does not; the schema is present or it is not.
- Directional: the brand was missing across several prompts today.
- Not Yet Measured: any revenue or retention impact from fixing the gap.
Honest labels keep a fast manual pass from turning into overclaiming.
Step 8: Turn findings into a build plan
End with a prioritized list, not a pile of screenshots.
For each confirmed gap, decide the action: publish an answer, correct a definition, add schema, surface proof, or connect internal links. That prioritized list is your build plan. For a deeper structural version of this pass, use How to audit your site against the SEO 3.0 stack, and remember why the monitoring alone is not the fix, covered in Prompt tracking does not fix AI search visibility.
Run the free EntityMesh scan after your manual teardown to compare your notes against a structured Authority Infrastructure diagnostic.
Frequently asked questions
How do I run an AI visibility teardown manually?
Choose 5 to 10 buyer-intent prompts, test them across a few AI systems, record brand mentions, competitor mentions, and cited sources, check description accuracy, compare results against your site's answer infrastructure, label the findings by confidence, and turn the gaps into a build plan.
How many prompts should I test?
For a 30-minute pass, 5 to 10 prompts across category, comparison, brand, and problem intents is enough to reveal patterns without overrunning the timer.
Which AI systems should I check?
Test more than one, such as ChatGPT, Perplexity, and Gemini, plus Google AI Overviews where relevant, because each system synthesizes from different sources.
What should I record during a teardown?
Record whether the brand appeared, its position in the answer, competitors mentioned, cited sources, description accuracy, and any definition drift, along with a saved copy of the raw output.
Can a manual teardown prove AI visibility improvement?
No. A manual teardown finds patterns to investigate. Prompt outputs vary, so a single test cannot prove causation or a stable trend; it is directional evidence that guides a build.
How does EntityMesh use teardown findings?
EntityMesh turns confirmed gaps into approved Support Hub, Answer Hub, FAQ, glossary, schema-ready, and internal-linking work, then uses EchoScan to monitor how the reflections change. It improves the conditions for citation but does not control external AI systems.