How to Run a Monthly AI-Visibility Audit (Free Prompt List + Template)

A monthly AI-visibility audit means running a structured set of prompts across ChatGPT, Perplexity, Google AI Overviews, and Claude, then scoring your brand on mention rate, citation accuracy, sentiment, and share of voice. It takes two to three hours without tooling. This guide gives you the exact prompts, a scoring template, and what to fix when the numbers are bad.

Author: Josh Bernstein, Co-Founder | Published: June 9, 2026

Why This Now Matters More Than Any Rank Report

Google search referral traffic to publishers fell 34% between December 2024 and December 2025, according to Chartbeat's cross-publisher dataset. At the same time, ChatGPT referral traffic to the open web more than doubled over the same period. The shift is structural, not cyclical.

The problem: most marketing teams still measure success via organic rank positions that increasingly do not translate to clicks. Google AI Overviews now appear in roughly 50% of all searches, and when one does appear, the zero-click rate climbs to 83%. Rank #3 in a query with an AI Overview means almost nothing if your brand is not cited inside that Overview.

AI-visibility audits exist to close the measurement gap. They answer one question: *when buyers ask an AI engine about your category, how often does your brand show up, how accurately, and what would have to change for it to show up more?*

If your team is doing generative engine optimization without a monthly audit loop, you are flying blind.

The Four Metrics That Actually Matter

Before running a single prompt, align your team on what you are measuring. Everything else is noise.

1. Brand Mention Rate

The percentage of your tracked prompts that surface your brand name in the AI response, anywhere, including passing references. This is your baseline visibility score.

2. Citation Rate

The subset of mentions where the AI engine links to or explicitly credits a specific page on your domain. A mention without a citation is weaker signal; a citation is a direct referral path. As a benchmark, the average brand mention rate across categories sits at 17.2%, per AthenaHQ's State of AI Search 2026 report, market leaders run 40-60%.

3. Sentiment and Accuracy Score

Whether the AI's description of your brand is positive, neutral, or negative, and whether it is factually correct. Inaccurate descriptions are an under-tracked risk: they can suppress purchase intent without you ever knowing the source.

4. Share of Voice (SoV)

Your brand mentions as a percentage of all competitor mentions across the same prompt set. This is the metric that maps most directly to commercial outcomes. Most marketing teams still have no systematic way to measure AI citations, even those who name AI search a core 2026 priority, and that gap is your competitive window.

Platform Differences You Cannot Ignore

Optimizing "for AI" as a monolith is a mistake. Each engine has distinct citation behavior that demands different inputs.

*Average citations per AI response by platform. Perplexity's high citation volume makes it easier to break into, but ChatGPT's lower, more selective set is harder to earn and carries more authority weight.*

Key implication: only 11% of domains cited by ChatGPT overlap with those cited by Perplexity. Running your audit on a single platform tells you almost nothing about your true cross-channel AI visibility. For a deeper breakdown of how each engine sources differently, see how AI engines cite brands differently.

Phase 1: Build Your Prompt Bank (30 Prompts, Three Types)

Your prompt bank is the foundation of every audit. It needs to cover three query types at roughly these ratios:

Category prompts (50% of bank, 15 prompts) These are the non-branded, buyer-intent queries where AI engines make recommendations. They reveal whether you appear as an option at all.

*Examples (swap in your category):*

• "What's the best [your category] tool for [use case]?"
• "Top platforms for [problem your product solves] in [year]"
• "Which [your category] tools do enterprise teams use?"
• "What should I look for in a [your category] solution?"
• "[Your category] tools compared"

Comparison prompts (25% of bank, 7-8 prompts) Buyer-intent queries where a decision is being made. Brand should appear and be described accurately.

*Examples:*

• "[Your Brand] vs [Competitor A] vs [Competitor B]"
• "Is [Your Brand] worth it for [use case]?"
• "What are the alternatives to [Your Brand]?"
• "Pros and cons of [Your Brand]"

Brand/direct prompts (25% of bank, 7-8 prompts) Queries where someone is specifically looking you up. These test accuracy and sentiment most directly.

*Examples:*

• "What is [Your Brand] and what does it do?"
• "How does [Your Brand] work?"
• "Who uses [Your Brand]?"
• "What do customers say about [Your Brand]?"

Tip: Run the full 30-prompt set on ChatGPT, Perplexity, and Google AI Overviews every month. Add Gemini and Claude quarterly. Do not rotate prompts mid-cycle, consistency is what makes month-over-month deltas meaningful.

Phase 2: Run the Audit and Score Each Response

For every prompt × platform combination, log five fields in your tracking sheet:

1. Mentioned? (Y / N)
2. Citation included? (Y / N, was a link to your domain in the response?)
3. Position in response (1st brand named / middle / end / absent)
4. Sentiment (Positive / Neutral / Negative)
5. Accuracy (Correct / Partially correct / Incorrect / Not described)

That is 150 data points for a 30-prompt × 3-platform audit. It sounds like a lot. In a spreadsheet with consistent formatting, a disciplined human can complete a full manual audit in about two hours. A platform like Promptive automates the run and scoring, useful once your prompt bank exceeds 30 queries or you are tracking more than three competitors.

Scoring formula

Calculate these four summary metrics from your raw data:

• Mention rate = (# prompts with brand mention) ÷ (total prompts) × 100
• Citation rate = (# prompts with citation to your domain) ÷ (total prompts) × 100
• SoV = (your brand mentions) ÷ (your brand mentions + all competitor mentions) × 100
• Accuracy rate = (# responses marked "Correct") ÷ (# responses with description) × 100

Phase 3: The Scoring Template

Use this table to track monthly results. Copy it into Sheets, Notion, or Airtable.

Benchmark targets (based on category leaders, 2026 data):

• Mention rate above 30%: competitive
• Mention rate above 50%: strong category presence
• SoV above 25%: market leader territory
• Citation rate above 20%: your content is being treated as authoritative

Phase 4: Diagnose the Gaps

*The audit loop is only valuable when it feeds a fix queue. One finding + one content action beats 30 findings with no owner.*

Different low scores point to different root causes:

Low mention rate, low SoV The AI engines have insufficient signal about your brand. Fix: publish more content in venues the engines trust, independent reviews, third-party round-ups, industry publications, and forum discussions. Your own domain is just one of many sources AI engines draw from, community platforms and earned media frequently dominate citation pools for informational and discovery queries. (How AI engines source citations) Publishing only on your own site is the single fastest way to stay invisible.

Present but not cited Your brand is mentioned but no link is pulled. Fix: strengthen your most authoritative pages, add structured headers, definition blocks, comparison tables, and clearly attributed data points. Claude consistently favors bullet-pointed, structured pages in its citation selection over unstructured prose. Google AIO favors Google-indexed publisher URLs. Know the platform you are optimizing for.

Cited but described inaccurately This is the most urgent fix. Wrong descriptions can suppress purchase intent at scale. Start with your own site: does your homepage and product page describe what you do in clear, unambiguous language? Does your Wikipedia entry (if you have one) reflect your current positioning? Are third-party review sites carrying outdated information? Each of these is a lever. Use the Promptive GEO audit tool to catch accuracy issues programmatically before they compound.

Good mention rate, low SoV Your brand appears but competitors dominate the response. This is a share-of-position problem, you need to move from middle-of-response mentions to being named first. Fix: create content that directly answers the prompts where you trail, and earn citations from higher-authority domains than the ones currently feeding the AI's response.

The 2-Hour Manual Audit Workflow

If you are running this without tooling, here is the exact sequence:

1. Open a fresh incognito or private session for each platform (prevents personalization bleed).
2. Copy your 30 prompts into a spreadsheet column. Columns: Platform | Prompt | Mentioned | Cited | Position | Sentiment | Accuracy.
3. Paste each prompt, copy the response, score it immediately, do not batch. Memory degrades fast across 30 responses.
4. Flag any response where accuracy is wrong and screenshot it. These are your highest-priority fix tickets.
5. Calculate your four summary metrics at the bottom of the sheet.
6. Compare to last month's baseline. A 5-point MoM improvement in mention rate is a meaningful positive signal; a 5-point drop warrants a content review to see what changed.

Total time at a steady pace: 90-120 minutes per full audit cycle.

For teams running audits on 10+ brands or tracking 5+ platforms, manual execution stops being practical around month two. The Promptive GEO audit tool runs this automatically and surfaces deltas without manual logging. For context on what the current AI search landscape looks like before you start auditing, the Q1 2026 state of LLM search report is the most current baseline available.

What Good Looks Like: Category Benchmarks

These figures are drawn from 2026 cross-category analysis. Use them to calibrate expectations, not as targets every brand should chase in month one.

Most B2B SaaS brands in competitive categories start somewhere in the 10-25% mention range on their first audit. That is not a crisis, it is a baseline. The glossary of GEO terms on Promptive defines each of these metrics in more detail if you need a shared vocabulary for stakeholder reporting.

Frequently Asked Questions

How often should I run an AI-visibility audit?

Monthly is the right cadence for your core 30-prompt set across your top three platforms. Run a deeper quarterly audit that expands prompts, adds platforms, and does a competitive gap review. Ad-hoc checks make sense after major content launches or any time you see a sudden change in AI referral traffic in your analytics.

What platforms should I include in a baseline audit?

Start with ChatGPT, Perplexity, and Google AI Overviews, these three together cover the majority of commercial AI search volume in 2026. Add Gemini and Claude in your quarterly deep audit. Platform behavior diverges enough that optimizing only for Google while ignoring Perplexity, or vice versa, leaves real gaps: only 11% of domains cited by ChatGPT overlap with Perplexity citations.

How many prompts do I need to get statistically meaningful results?

30 prompts across three platforms (90 data points per month) gives you enough signal to detect meaningful trends. Fewer than 20 prompts and your mention-rate percentage becomes too volatile, a single response change swings your score by 5+ points. More than 50 prompts is valuable but raises execution cost; use tooling if you go above that threshold.

My brand has a zero percent mention rate. Where do I start?

Focus on earned media first. Distributing your content to reputable third-party publications consistently increases AI citations, earned placements on high-authority domains carry substantially more weight than publishing on your own domain alone. Identify the two or three publications that consistently appear in AI responses for your category, and earn placements there. Then structure your own site's most important pages with clear headers, definition blocks, and attributed statistics.

How is AI share of voice different from traditional search share of voice?

Traditional SoV counts ranking positions and estimated impressions. AI SoV counts how often your brand is named across a defined prompt set relative to named competitors. The key difference: in traditional search you can rank #1 without a competitor being mentioned; in AI search, a response that mentions four competitors and not you is a zero for your brand regardless of your organic rankings.

Can I trust AI responses to be consistent from one run to the next?

No, AI engines are probabilistic. The same prompt can return different responses on two consecutive runs. That is why you run each prompt once per month rather than comparing hourly or daily. Monthly averages smooth the stochasticity enough to reveal real trends. If you want higher confidence, run each prompt three times per session and score the majority result.