AI visibility audits with Claude Code: competitor analysis and content gaps

TL;DR

• Traditional SEO audits measure rankings and backlinks. AI visibility audits measure citation rate and share of voice across ChatGPT, Claude, Perplexity, and Gemini, which are the surfaces where B2B buyers now research vendors.
• Claude Code lets you programmatically test hundreds of high-intent buyer queries, extract citation data into structured CSV output, and benchmark competitor mention rates without hours of manual copying.
• Our analysis shows only 38% of AI Overview citations now come from top-10 organic results, down from 76% in mid-2025.
• The CITABLE framework provides a structured method to restructure content for passage retrieval and close citation gaps after you identify them.
• The four plug-and-play prompts in this guide automate citation extraction, competitor benchmarking, intent gap analysis, and content prioritization, making the workflow executable for a lean marketing team.

Stable organic rankings and strong AI citation rates rarely move together, and for most B2B SaaS brands, the gap is widest on the category queries buyers use to evaluate vendors. Fixing that requires a different kind of audit. Traditional SEO tools measure keyword positions and backlink counts. AI visibility audits measure something else: how often your brand appears in synthesized LLM answers, which third-party sources drive those citations, and which queries competitors own that you don't. This guide walks you through how to build that audit pipeline using Claude Code, map your citation gaps against competitors, and generate a prioritized content roadmap using the CITABLE framework.

How AI audits reveal hidden growth levers

An AI visibility audit maps your brand's presence across the three surfaces where modern organic search operates: web search, AI citations, and training data. Traditional SEO audits only address the first. The citation surface is where B2B buyers evaluate vendors inside ChatGPT, Claude, and Perplexity without clicking any link, and the training data surface shapes which brand associations LLMs carry before they run a search query.

The audit can help identify which specific queries trigger competitor citations instead of yours, and which third-party sources (such as Reddit threads, G2 reviews, analyst write-ups) may be feeding those citations. With that data, you build a content and off-page plan that targets gaps directly rather than optimizing for signals that no longer drive AI visibility. We document how this plays out in client engagements in our B2B SaaS SEO case studies.

Why benchmarking reveals content gaps

Benchmarking your AI citation rate against direct competitors surfaces the gap and explains the mechanism faster than auditing in isolation. Traditional keyword gap analysis shows which terms competitors rank for that you don't. AI citation gap analysis shows which buyer queries produce competitor citations instead of yours, and it identifies why: information consistency and third-party validation appear to drive selection in a Retrieval-Augmented Generation (RAG) system, not link authority alone.

Our Answer Engine Optimization (AEO) expertise guide covers this distinction in more depth, including why generalist SEO agencies struggle to execute against the retrieval mechanics involved.

Defining your AI visibility footprint

Your AI visibility footprint is the set of buyer queries where your brand appears in LLM-synthesized answers across ChatGPT, Claude, Gemini, and Perplexity. Two metrics define it. Citation rate is how often AI models link to or explicitly cite your content as a source in the tested query set. Mention rate tracks how frequently your brand name appears in generated answers.

Mapping your footprint before any intervention gives you the baseline that makes progress defensible to a CFO or board. Our three surfaces guide explains how web search, citations, and training data each require different tactics, and why a single metric like organic traffic obscures most of what's actually happening.

Building your AI visibility audit workflow

The audit pipeline we describe runs in five stages. Claude Code can automate large portions of the middle stages, converting what would be many hours of manual testing into a structured, repeatable process you can run quarterly.

[Query Universe] ---> [Claude Code[Code / API Testing] ---> [Citation Aggregation] ---> [Gap Analysis] ---> [Prioritized Action List]

Each stage builds on the last. The query universe defines scope. Claude Code can help automate portions of the testing workflow by calling LLM APIs programmatically. Citation aggregation structures the raw output. Gap analysis identifies where competitors appear and you don't. The prioritized action list maps those gaps to specific content assets. Our Claude Code for AEO workflows guide covers the general architecture for this kind of pipeline.

Building your AI query universe

Start by building a query universe of high-intent buyer questions for your B2B SaaS category. A working set of 100-200 queries is a reasonable starting point. Focus on queries that reflect actual research behavior: "best [category] for [use case]," "how to [solve problem] with [category]," "[your category] compared to [competitor category]." These represent the highest pipeline value per citation because they map to active vendor evaluation.

Group queries by buyer-research intent: informational (how does X work), comparative (X vs Y), commercial (best X for Z), and navigational (brand-specific searches). This grouping typically de-prioritizes navigational queries for competitive analysis, since they rarely surface competitor citations in meaningful volumes. You'll use these categories later to identify which buyer journey stage competitors dominate. Our competitor gap analysis guide covers query universe construction in detail.

Testing visibility across AI surfaces

Different LLMs weight sources differently, so testing against a single platform understates your real visibility gap. Run your query set across ChatGPT (via the OpenAI API), Claude (via Anthropic's API), and Perplexity at minimum. For B2B categories with high enterprise buyer overlap, Google AI Overviews and Microsoft Copilot can provide additional data points.

Two LLMs can return entirely different citations for the same query because their retrieval mechanisms weight different signals. Testing across surfaces gives you a realistic share of voice picture rather than an optimistic one.

Benchmarking competitor citation sources

When an LLM cites a competitor instead of you, it draws from a specific third-party source. Identifying those sources matters as much as counting citations. Run your competitor names as entities through the same query set and extract the citation URLs from each response.

Our research across 144,000 AI citations found Reddit captured 0.35% of all visible ChatGPT citations but occupied a substantial portion of ChatGPT's internal search slots during query processing. That divergence means tools that only count surfaced URLs miss a large share of what's actually shaping AI answers. When competitors show up in LLM responses, Reddit threads, G2 reviews, and independent comparison articles often appear as underlying sources.

Finding untapped AI citation wins

Not all gaps are defensive. Some queries have no dominant competitor citation, either because the category is underserved on that question or because the LLM is synthesizing from fragmented, low-quality sources. These represent open opportunities to establish authority quickly.

In your extracted data, look for queries where the LLM generates a generic answer without citing a specific brand, or where it cites older, less authoritative content. LLMs using RAG architecture retrieve from the live web, so fresher content with explicit timestamps can displace older citations over time. A well-structured, recent asset can move into citation position after publication and indexing, though timelines vary.

Structuring your AI optimization roadmap

Translate the gap analysis into a prioritized execution plan organized by impact. High-impact gaps are queries where competitors are strongly cited and the opportunity is large. Medium-impact gaps are queries with fragmented citations and no clear competitor authority: these can often be addressed with a single optimized asset built to the CITABLE framework's block-structured format. Lower-impact opportunities are queries where you have some citation presence but rank below competitors: these are optimization jobs, not creation jobs.

How to automate your AI visibility audit

Manual query testing doesn't scale well without consuming significant team time. Claude Code can help solve this. It's a coding tool that reads files, runs scripts, calls APIs, and outputs structured data, all within your terminal or IDE. You write the workflow once and it runs programmatically. Full setup is covered in our Claude Code installation guide.

Building your query universe

Feed Claude Code your seed keyword list and instruct it to generate a semantic query set. A seed list of category terms can expand to a larger set of distinct buyer queries when Claude accounts for intent variants, use-case modifiers, and comparison formulations.

Use a command like this to generate your query universe:

/generate-query-universe
Seed terms: [your 10-15 category keywords]
Output format: CSV with columns: query | intent_tier | priority_score
Intent tiers: informational, comparative, transactional
Priority scoring: weight transactional queries highest

Claude Code can read the seed list from a text file in your project directory and output a structured CSV you can review before running the full API test cycle. Our skill library guide shows how to make these commands reusable across audits.

Extracting AI responses at scale

With your query universe ready, an extraction script can call LLM APIs, parse responses, and write citation data to a structured output file. This example assumes you've defined extract_brand_mentions() and extract_urls() helper functions to parse response text:

import anthropic
import csv

queries = open("queries.txt").readlines()
brand_list = ["YourBrand", "CompetitorA", "CompetitorB"]
results = []

for query in queries:
    client = anthropic.Anthropic()
    message = client.messages.create(
        model="claude-3-5-sonnet-20241022",
        max_tokens=2048,
        messages=[{"role": "user", "content": query.strip()}]
    )
    response_text = message.content[0].text

    results.append({
        "query": query.strip(),
        "platform": "claude",
        "response": response_text,
        "cited_brands": extract_brand_mentions(response_text, brand_list),
        "cited_urls": extract_urls(response_text)
    })

with open("citation_data.csv", "w") as f:
    writer = csv.DictWriter(f, fieldnames=["query","platform","response","cited_brands","cited_urls"])
    writer.writeheader()
    writer.writerows(results)

Run the same loop against the OpenAI API for ChatGPT data and against the Perplexity API for a third surface. The output CSV becomes the foundation for every subsequent analysis step.

Quantifying your AI citation gap vs competitors

Raw LLM responses are qualitative. To present audit findings to a board or CFO, you need numbers. This section converts the extracted CSV data into three metrics: citation rate, AI share of voice, and intent category coverage, each of which maps directly to a pipeline conversation.

Assessing competitor citation footprints

Load your citation_data.csv and filter by brand mentions. For each competitor, count the queries where that brand appears in the LLM response. That count divided by total queries tested gives their citation rate. Do the same for your own brand.

The citation URL column tells you which external sources drive those mentions. Sorting cited URLs by frequency can reveal the domains where competitors appear most consistently. Google's AGREE research describes how LLMs ground claims by referencing documents, which means those frequently-cited domains are specific places where building your own presence may strengthen your citation potential.

Measuring competitor AI share of voice

AI share of voice is the percentage of total brand citations your brand wins across the full query set. Calculate it with this formula:

(Your Brand Citations / Total Citations Across All Brands) x 100

If your brand appears in 42 query responses with citations and competitors collectively generate 78 citations across the same set, your share of voice is 42 / (42+78) x 100 = 35%. That number is the most defensible metric to bring into a board review because it's relative, directional, and moves as your content program delivers. Our AEO performance metrics guide covers how to track share of voice alongside citation rate over time.

Benchmarking leaders via intent categories

Group your citation data by intent tier. Calculate share of voice separately for informational, comparative, and transactional queries. Most B2B SaaS brands find some presence in informational queries because they've published blog content, but are largely absent in comparative and transactional queries because they haven't structured content to answer "X vs Y" or "best X for Z" questions directly.

Our citation analysis shows that comparative and transactional queries map directly to buyer decision points, making them the highest pipeline-value gaps to close.

Finding gaps in your buyer evaluation process

AI search changes the buying process by moving the consideration phase inside the LLM interface. A prospect who asks Claude "best incident management platform for engineering teams" and gets three competitor names cited has already formed a consideration set without visiting your site. That interaction is invisible to GA4, HubSpot, and Salesforce.

Analyzing competitor citation gaps

From your intent-segmented data, pull every query where at least one competitor is cited and your brand is not. This is your citation gap list. Sort it by intent tier, with transactional queries at the top. For each gap query, note the citation source URL: this tells you whether the competitor wins from their own content, a third-party review, or a community discussion.

Our own tracking shows top-10 organic rankings accounted for 76% of AI Overview citations in mid-2025, falling to 38% by early 2026. External tracking by almcorp puts some category figures as low as 17%. Either way, ranking #1 on Google for a query is no longer a reliable proxy for winning its AI citation. The citation gap analysis you build here can help identify opportunities that ranking reports may miss.

Uncovering untapped AI search opportunities

Scan for queries where no competitor wins a clean citation: the LLM returns a generic synthesis with no named brand or a mix of partially-relevant mentions. These represent open territory where a single well-structured asset can establish authority before any competitor moves. They appear most often in niche use-case queries and in emerging integration categories where no one has published direct-answer content.

Publishing a direct-answer piece on a query where competitors win with older content can be a high-yield move with a short payback window, since RAG systems retrieve from the live web and fresher content with accurate facts can displace older citations.

Connecting gap queries to your buying cycle

Overlay your gap list against your product's specific use cases and differentiators. A query like "best [category] for enterprise security teams" typically signals stronger pipeline intent than "what is [category]" because it reflects a buyer who is evaluating, not just learning. Map each gap query to the stage of your buying cycle it represents: awareness, consideration, or decision. This mapping becomes the filter that determines which gaps you close first when building your content queue.

Prioritizing content for maximum AI impact

With a full gap list segmented by intent and buyer stage, prioritization separates high-yield opportunities from noise. You want to close the gaps that sit at evaluation and decision stages and where competitors are already building citation consistency. Those are the queries that directly influence pipeline.

Filtering your gap list by buyer intent

Filter your gap list to keep only queries with comparative or commercial intent at the consideration or decision stage. This typically narrows your list to a more focused set of high-priority targets. These are the queries where a missing citation can cost you consideration set inclusion, not just awareness. The CITABLE framework's Intent architecture component is specifically designed to ensure each asset answers the main query and the adjacent questions buyers have at that stage.

Analyzing competitor citation sources

For each high-priority gap query, note the competitor citation source URLs and cross-reference them with your own off-page presence. If a competitor is being cited because a Reddit thread consistently recommends them and your brand has no presence in that subreddit, that's an off-page action item, not a content writing job. LLMs appear to weight citation selection based on a combination of semantic relevance and information consistency across independent sources, though the exact retrieval mechanics vary by model and are not fully documented publicly.

Our Reddit and ChatGPT research found Reddit captured 0.35% of visible ChatGPT citations, making it a high-value off-page surface for citation consistency work.

Prioritizing your high-impact content queue

Build a content queue ranked by: intent tier (transactional first), competitor citation strength in the gap (close gaps where competitors are strongly cited before open-field opportunities), and your product's ability to directly answer the query. Assign each item a CITABLE framework checklist: does the existing asset (if one exists) have a BLUF opening, block-structured sections, third-party validation signals, and schema markup?

Our retainer programs can deliver CITABLE-optimized articles with dedicated team support covering content, SEO, off-page consistency, and structured data. See our AEO agency vs in-house breakdown if you're weighing internal execution against outsourcing.

4 plug-and-play prompts for audit workflows

These four prompts are ready to run inside Claude Code. Each targets a specific audit stage. Copy them into your .claude/commands/ directory as Markdown files and invoke them with a slash command.

Prompt 1: Gather AI citation samples

Use this prompt to extract citations and source URLs from raw LLM responses stored in a CSV file.

You are an AI citation extraction specialist.
I will provide a CSV file where each row contains:
- Column A: the query tested
- Column B: the raw LLM response text

Your task:
1. Identify every brand name mentioned in each response.
2. Extract every URL cited.
3. Classify each mention type (positive, neutral, or comparative).
4. Output a new CSV with sample columns such as:
   query | brand_mentioned | mention_type | cited_url | citation_position

File path: ./raw_responses.csv
Output path: ./citation_samples.csv

Prompt 2: Benchmark rival citation rates

Use this prompt to calculate share of voice and citation frequency across a competitor set.

You are an AI share of voice analyst.
I will provide a CSV file with columns such as:
query | brand_mentioned | mention_type | cited_url | citation_position

My brand is: [YOUR BRAND]
Competitors to benchmark: [COMPETITOR A], [COMPETITOR B], [COMPETITOR C]

Your task:
1. Count total citations per brand across all queries.
2. Calculate citation rate: (queries with brand mention / total queries) * 100.
3. Calculate AI share of voice: (brand citations / total citations) * 100.
4. Output example columns: brand | citation_count | citation_rate | share_of_voice_pct

File path: ./citation_samples.csv
Output path: ./sov_benchmark.csv

Prompt 3: Analyzing competitor intent gaps

Use this prompt to identify which buyer evaluation stages competitors dominate and where your brand is absent.

You are a buyer journeyevaluation stage intent analyst.
Files provided:
1. ./query_universe.csv: query | intent_tier | priority_score
2. ./citation_samples.csv: citation data per query and brand

My brand is: [YOUR BRAND]

Your task:
1. Merge files using the query field.
2. For each intent tier, show: top competitor citation rate,
   my brand's citation rate, and the gap between them.
3. Flag high-priority queries (such as top 10 transactional queries)
   where my brand has zero citations and competitors have strong presence.
4. Output: ./intent_gap_analysis.csv
   Example columns: intent_tier | query | top_competitor |
   competitor_citation_rate | my_citation_rate | gap_score

Prompt 4: Analyze AI citation gaps

Use this prompt to map missing citations to specific content requirements using the CITABLE framework.

You are an AEO content strategist applying the CITABLE framework.
CITABLE: Clear entity and structure, Intent architecture,
Third-party validation, Answer grounding, Block-structured for RAG,
Latest and consistent, Entity graph and schema.

Input: ./intent_gap_analysis.csv
Focus: rows with significant gaps in high-intent categories

For each flagged query:
1. Identify missing CITABLE components based on query type.
2. Recommend asset type: new article, page restructure,
   or off-page consistency action.
3. Estimate typical time-to-citation based on asset complexity.
4. Output: ./content_priority_queue.csv
   Example columns: query | gap_score | citable_gap | asset_type |
   estimated_timeline | recommended_action

Illustrative example: SaaS AI visibility auditing

The following is an illustrative example based on patterns we see across client engagements, not a single named project. A collaborative workspace SaaS platform with stable Google rankings but limited AI-referred pipeline presents a common audit starting point. They reported seeing trials from AI recommendations, while top competitors were consistently cited in LLM responses to their highest-value buyer queries. We ran an audit using the pipeline described above across high-intent queries.

Grouping queries by buyer intent

The query universe focused on high-intent terms like "best collaborative document editor for engineering teams," "real-time collaboration tools compared," and "how to manage technical documentation for distributed teams." We grouped them by buyer-research intent, weighting comparative and commercial queries heavily because those map directly to the consideration phase.

The query mapping revealed a structural blind spot: the client had published strong informational content but had limited assets directly addressing comparative buyer queries. That content gap appeared to explain a significant portion of the citation absence.

Where competitors outperform us

The audit showed competitors dominated citations across the comparative and transactional query tiers. Source analysis identified why: competitors appeared to have consistent product claims across Reddit threads in target subreddits, G2 comparison pages, and independent developer blogs. The client's brand appeared in some of these sources but with fragmented claims and inconsistent positioning, which likely meant LLMs encountered conflicting signals and defaulted to competitors with cleaner information consistency.

The workspace SaaS faced the same issue: presence without consistency is not enough for RAG systems.

Targeting high-impact citation gaps

We restructured core assets using the CITABLE framework, prioritizing the B-component (block-structured for RAG: sections with bottom-line-up-front openings, explicit entity naming) and third-party validation components (coordinated presence in relevant communities and review platforms). We also worked to standardize product claims across external sources to resolve conflicting consistency signals. The result across the broader engagement was significant growth in AI-referred trials. Our full methodology walkthrough covers the execution sequence in detail.

Clarifying AI visibility audit objectives

An audit is only as useful as the expectations around it. Setting realistic timelines and sample sizes before you start means the findings land as action items, not abstract data.

Audit sprint length and milestones

A manual audit can be time-intensive. Automation through Claude Code and API scripts significantly reduces the effort required. Our AEO Sprint (€6,995 one-off) delivers a programmatic audit across all major LLMs, a competitor share-of-voice benchmark, schema optimization across priority pages, an entity map, and CITABLE-optimized articles. For teams evaluating the ROI model before committing to a retainer, our AEO payback calculator walks through the pipeline math.

Determining optimal query sample sizes

Testing 100-200 high-intent queries is a reasonable range for most B2B SaaS audits. Fewer queries risk missing key competitive patterns, while substantially larger sets can add API cost without proportional signal improvement for mid-market categories. For enterprise SaaS with broader category coverage, a larger query set split across multiple intent categories can provide the statistical depth to make the findings defensible to stakeholders.

Setting your re-audit schedule

Re-audit quarterly. LLMs update frequently, competitor content ships continuously, and the overlap between organic rankings and AI citations is still shifting, dropping from 76% to 38% in roughly six to nine months per our own tracking. A quarterly re-audit catches model behavior shifts before they produce visible pipeline drops. Run citation tracking using your extraction scripts between full audits to maintain a directional read on share of voice momentum.

Automating citation data collection

Our AI Visibility Tracker can automate ongoing citation collection, tracking citation rate, share of voice, and competitor footprint across ChatGPT, Claude, Perplexity, and Gemini, and feeds that data back into recommendations for what to ship next. For teams building their own tooling first, the Claude Code workflow in this guide covers the same measurement logic.

The audit pipeline in this guide gives you the data. The CITABLE framework gives you the content structure to act on it. Combined, they turn the question your CEO is asking ("why are competitors cited and we're not?") into a documented, measurable plan with visible month-on-month progress. If you want us to run this audit for you, the AEO Sprint is the right starting point. Book a call and we'll tell you honestly whether we're a fit.

FAQs

How long does an AI visibility audit take?

A manual audit can be time-intensive. Our AEO Sprint delivers a complete programmatic audit across all major LLMs, a competitor share-of-voice benchmark, and prioritized recommendations at a one-off cost of €6,995.

What is a good AI citation rate for B2B SaaS?

Most unoptimized B2B SaaS brands see very low citation rates across their category query set at baseline. Structured CITABLE optimization and off-page consistency work typically produces measurable citation improvements within the first few months, with initial citations often appearing within one to two weeks.

How much does a professional AI visibility audit cost?

Our one-off AEO Sprint costs €6,995 and includes a full programmatic audit across ChatGPT, Claude, Gemini, and Perplexity, competitor share-of-voice benchmarking, schema optimization, an entity map, and CITABLE-optimized articles.

How often should I re-run an AI visibility audit?

Quarterly re-audits are a recommended cadence for most B2B SaaS teams, with regular citation tracking between full audits. Our own tracking shows the overlap between top-10 organic rankings and AI citations dropped from 76% in mid-2025 to 38% by early 2026, meaning the citation environment shifts fast enough that annual audits miss meaningful competitive movement.

What is AI share of voice and how is it calculated?

AI share of voice is the percentage of total brand citations your brand wins across a defined set of category queries, calculated as (your brand citations / total citations across all brands tested) x 100. It's the primary metric for presenting AI visibility progress to a board because it's relative to the competitive set and directional over time.

Can I run this audit without Claude Code?

Yes. You can manually test queries across ChatGPT, Claude, and Perplexity and record results in a spreadsheet. Manual testing is more time-intensive than the automated workflow, but the methodology is identical.

Key terms glossary

AI share of voice (SoV): The percentage of total brand citations your brand wins across a specific set of category queries, calculated as (brand citations / total citations) x 100.

Citation rate: How often AI models link to or explicitly cite your content as a source across a tested query set.

Information consistency: The alignment of product claims and facts across independent sources (Reddit, G2, industry sites, your own content), which appears to influence citation selection in LLM systems.

Mention rate: The frequency of brand name appearances within generated LLM answers, tracked separately from citation presence.

Passage retrieval: The mechanism where an LLM extracts a specific block of semantically relevant text to answer a query, independent of page-level ranking signals.

RAG (Retrieval-Augmented Generation): The architecture where LLMs retrieve external information during answer generation, rather than relying solely on training data.