Key takeaways
- A GEO content audit has two distinct phases: diagnosis (finding why AI models skip your content) and action (fixing it). Most tools only do the first.
- AI engines cite passages, not pages — so every H2 section needs to work as a standalone, self-contained answer.
- Crawl errors, JavaScript rendering issues, and missing schema are the most common technical blockers that prevent AI crawlers from reading your content at all.
- Answer gap analysis — seeing which prompts your competitors rank for but you don't — is the highest-leverage starting point for new content.
- Tracking the full loop (crawl → citation → traffic) is the only way to know whether your fixes are actually working.
Running a GEO audit sounds straightforward until you're three hours in and staring at a spreadsheet of 400 URLs with no clear idea what to fix first. The problem isn't a lack of data. It's that most auditing workflows are built around diagnosis, not action. You find out you're invisible in ChatGPT and Perplexity. Then what?
This guide walks through a complete GEO content audit process — from the initial crawl through to published fixes — using tools designed to close that loop. It's structured around the actual sequence of work, not a theoretical framework.
Phase 1: Inventory — know what you actually have
Before you can optimize anything, you need an accurate map of your site. This sounds obvious, but most teams underestimate how messy their content inventory really is. Thin tag archives, paginated blog indexes, orphaned landing pages from campaigns three years ago — these all show up in a crawl and need to be classified before you can make any decisions.
Running the crawl
For sites under 10,000 URLs, Screaming Frog SEO Spider is the standard starting point. It's fast, configurable, and exports clean CSVs.
For larger or JavaScript-heavy sites, you'll need something that handles dynamic rendering. Firecrawl is built specifically for this — it's API-based and designed for AI and SEO workflows where you need clean, structured output rather than raw HTML.
One thing to check immediately: whether your crawler is respecting robots.txt in a way that's hiding content. AI crawlers like GPTBot and ClaudeBot follow their own rules, and a misconfigured robots.txt can silently block them from your most important pages. You won't know this is happening unless you're looking at actual crawler logs.
Classifying your URLs
Once you have the crawl data, classify every URL by template type: home, pillar, blog post, product, category, landing page, about, contact, legal, utility. This classification drives every subsequent decision — which pages to audit for chunk readiness, which need schema, which are worth fixing at all.
Flag anything where the template is ambiguous. A page that looks like a blog post but has product pricing buried in it needs a different treatment than a standard editorial article.
Export this as a CSV. It's the artifact everything else references.
Phase 2: Chunk audit — can AI actually quote your content?
This is where GEO auditing diverges from traditional SEO. Search engines rank pages. AI engines quote passages. That distinction changes everything about how you evaluate content quality.
A "chunk" is any H2 section that could be lifted out of its page and still make complete sense. No pronoun dependencies. No "as we mentioned above." No heading like "The next step" that requires the surrounding context to be meaningful.
Scoring chunk readiness
Score each H2 section on a simple 0-1-2 rubric:
| Score | Meaning |
|---|---|
| 0 | No direct-answer opening sentence, or no H2 structure at all |
| 1 | Structured, but first sentence uses pronouns or references earlier content |
| 2 | Self-contained: opens with a direct answer, no external dependencies |
Sample 10-25% of URLs per template. Cover all pillar pages fully. For blog posts, a representative sample is enough to identify patterns.
Calculate a page-level average and a site-level median. The median tells you where your baseline is. Pages scoring below 1.0 on average are essentially invisible to AI engines regardless of how well they rank in Google.
What fails most often
From audits across B2B sites, the most common chunk failures are:
- Sections that open with "This is important because..." instead of the actual answer
- H2 headings that are vague labels ("Overview", "Key considerations") rather than answerable questions
- Content that buries the conclusion in the final paragraph after three paragraphs of context-setting
- Lists that make sense only because of the introductory sentence above them
Fix these and you've addressed the majority of why AI models skip your content.
Phase 3: Technical readiness — can AI crawlers reach your content at all?
A perfectly written page that AI crawlers can't access is worthless. Technical GEO readiness covers four areas: crawlability, rendering, schema, and page speed.
Crawlability and rendering
Check your robots.txt for rules that block GPTBot, ClaudeBot, PerplexityBot, or Googlebot. Many sites have legacy rules that were added to block scrapers and inadvertently block AI crawlers too.
JavaScript rendering is a bigger problem than most teams realize. If your content is loaded client-side, AI crawlers may see an empty shell. Tools like Prerender.io and similar pre-rendering services solve this by serving static HTML to bots.
Schema markup
Schema is how you tell AI engines what type of content a page contains, who wrote it, when it was published, and what questions it answers. The highest-value schema types for GEO are:
ArticleorBlogPostingwithdatePublished,dateModified,author, anddescriptionFAQPagefor pages with question-and-answer sectionsHowTofor step-by-step guidesOrganizationandBreadcrumbListfor site-level entity signals
WordLift is worth mentioning here — it's specifically built around structured data and entity optimization, which is directly relevant to how AI engines understand and cite content.
Page speed
Slow pages get crawled less frequently. Google PageSpeed Insights gives you the baseline data, and GTmetrix shows you the specific bottlenecks.
Phase 4: Answer gap analysis — find what's missing
This is where the audit shifts from diagnosis to strategy. Answer gap analysis asks: which prompts are your competitors getting cited for that you're not?
This is different from keyword gap analysis. You're not looking at search rankings. You're looking at which questions AI engines answer using competitor content instead of yours — and why.
How to run it
Start by defining your "money prompts" — the 20-50 questions your target buyers are most likely to ask AI engines when evaluating your category. These aren't just keywords. They're full questions: "What's the best [product type] for [use case]?" or "How does [your category] work?"
Then test each prompt across ChatGPT, Perplexity, Claude, and Google AI Overviews. Record:
- Whether your brand appears
- Which competitors appear
- Which specific pages or sources are cited
- What the response says about your category
The gaps — prompts where competitors appear and you don't — become your content brief backlog.
Promptwatch automates this at scale. Its Answer Gap Analysis surfaces exactly which prompts competitors rank for but you don't, with prompt volume estimates and difficulty scores so you can prioritize. Rather than manually testing prompts one by one, you get a ranked list of gaps with the specific content your site is missing.
For teams that want to track this manually or with lighter tooling, tools like Otterly.AI and Peec AI provide basic prompt monitoring.
Otterly.AI
The difference is that monitoring tools show you the gap. An action platform helps you close it.
Phase 5: Competitor and citation analysis
Knowing you're invisible is one thing. Understanding why competitors are getting cited is more useful.
For each gap prompt, look at:
- Which specific pages competitors are getting cited for (not just their domain)
- What those pages do structurally that yours don't (direct answers, FAQ sections, schema, specific data points)
- Which third-party sources — Reddit threads, YouTube videos, review sites — are also being cited
That last point matters more than most teams expect. AI engines don't just cite brand websites. They cite Reddit discussions, YouTube tutorials, comparison articles on third-party sites, and review platforms. If your competitors are winning citations partly because they have active Reddit presence or well-cited YouTube content, that's a different fix than rewriting your own pages.
Ahrefs is useful here for backlink and citation pattern analysis at the domain level.
For AI-specific citation tracking — seeing exactly which pages, Reddit posts, and external sources are driving citations in LLM responses — Promptwatch's citation and source analysis goes deeper than traditional SEO tools.
Phase 6: Entity and authority signals
AI engines build a model of who you are as an entity — your brand, your expertise, your relationships to other entities. Weak entity signals mean AI models are less confident citing you, even when your content is technically well-structured.
What to check
- Is your brand consistently named the same way across your site, your schema, and external mentions?
- Do you have an
Organizationschema withsameAslinks to your LinkedIn, Crunchbase, Wikipedia (if applicable), and other authoritative profiles? - Are your authors identified with
Personschema and linked to their credentials? - Are you cited by sources that AI engines already trust?
Entity consistency is surprisingly easy to break. A company that goes by "Acme Corp" on its website but "Acme Corporation" in press releases and "ACME" in its Twitter bio creates ambiguity that makes AI engines less likely to confidently recommend it.
SE Ranking has solid on-page and schema auditing features that can help surface these inconsistencies.
Phase 7: Building the fix plan and generating content
This is where most GEO audits fall apart. You have a spreadsheet of gaps, a list of pages that need rewriting, and a schema audit with 40 items. Without a prioritized action plan, the whole thing sits in a folder.
Prioritization framework
Score each fix on two dimensions: impact (how much visibility could this unlock?) and effort (how long will this take?).
| Fix type | Typical impact | Typical effort |
|---|---|---|
| Add FAQ schema to existing pages | Medium | Low |
| Rewrite H2 sections for chunk readiness | High | Medium |
| Create new content for answer gaps | High | High |
| Fix robots.txt blocking AI crawlers | High | Low |
| Add author schema and credentials | Medium | Low |
| Build external citations (Reddit, YouTube) | High | High |
Start with the high-impact, low-effort fixes. Unblocking AI crawlers and adding schema to existing pages can move visibility scores within weeks.
Generating content for gaps
For new content targeting answer gaps, the brief needs to be grounded in actual prompt data — not just keyword research. What question is being asked? What does the current top-cited answer look like? What's missing from it that you could answer better?
Promptwatch's Content Agents generate articles and briefs directly from gap analysis data, using real prompt volumes, competitor citation patterns, and brand guidance. The output is content engineered to answer specific gaps, not generic SEO filler.
For content production at scale, tools like Jasper and AirOps are worth evaluating — both are built around structured content workflows rather than one-off generation.
For SEO-grounded content briefs, MarketMuse and Frase both do solid work on topic modeling and competitive content analysis.
Phase 8: Tracking results — from publish to citation
Publishing content is not the end of the audit. It's the beginning of the measurement phase.
What to track
After publishing a fix or new piece of content, you want to know:
- When did AI crawlers first visit the page?
- When did the page start appearing in AI citations?
- Which models are citing it, and for which prompts?
- Is that citation driving actual traffic?
This timeline — from publish to crawl to citation to traffic — is what tells you whether your GEO strategy is working. Without it, you're flying blind.
Most monitoring tools don't track this sequence. They show you current visibility but not the trajectory. Promptwatch's Agent Analytics logs AI crawler visits in real time, shows when pages move from crawled to cited, and connects citation visibility to traffic attribution. That's the feedback loop that turns a one-time audit into a continuous optimization process.
Setting realistic timelines
AI citation timelines are slower than Google indexing. A new page might get crawled by GPTBot within days, but it can take 4-8 weeks before it starts appearing in ChatGPT responses. Schema and structural fixes on existing pages tend to show results faster than brand-new content.
Track weekly, not daily. The signal-to-noise ratio on daily tracking is too low to be useful.
Putting it all together: the audit workflow
Here's the complete sequence as a repeatable process:
- Crawl and classify — full site inventory, URL classification, orphan detection
- Chunk audit — score H2 sections for self-contained citability, identify failure patterns by template
- Technical audit — robots.txt, rendering, schema, page speed
- Answer gap analysis — define money prompts, test across AI engines, map competitor citations
- Competitor and citation analysis — understand why competitors are cited, including offsite sources
- Entity audit — brand consistency, author schema, external authority signals
- Prioritized fix plan — rank by impact/effort, start with quick wins
- Content generation — briefs and articles grounded in gap data
- Publish and track — monitor crawl logs, citation timelines, and traffic attribution
The teams that get the most out of this process treat it as a cycle, not a one-time project. Run the audit, publish fixes, track results for 6-8 weeks, then run the gap analysis again. Your competitors are publishing content too. The gaps shift.
Tool comparison: monitoring vs. action platforms
| Tool | Prompt monitoring | Crawl logs | Gap analysis | Content generation | Traffic attribution |
|---|---|---|---|---|---|
| Promptwatch | Yes | Yes | Yes | Yes | Yes |
| Otterly.AI | Yes | No | No | No | No |
| Peec AI | Yes | No | No | No | No |
| AthenaHQ | Yes | No | Limited | No | No |
| Profound | Yes | No | Limited | No | No |
| Ahrefs | Partial | No | No | No | No |
| Semrush | Partial | No | No | No | No |
The pattern is consistent: most tools in this category are monitoring dashboards. They show you where you stand. An action platform shows you where you stand and helps you move.
Final thought
The GEO audit is only as valuable as what happens after it. A 40-page audit report that sits in a shared drive doesn't improve your AI visibility. The teams winning in AI search right now are the ones who've built a repeatable loop: find the gaps, create the content, track the citations, repeat.
The tools exist to do this efficiently. The question is whether your workflow is built around diagnosis or action.