
ℹ️ Geodeck is built by the team behind Seofable, an AI-era SEO content tool. Seofable is listed in our directory as a clearly labeled featured listing; rankings and recommendations in this article are editorial.
What Is an AI Visibility Audit (and What It Isn't)
An AI visibility audit is a structured check of whether a brand is discoverable, extractable and citable inside AI-generated answers. It measures three separate things: does ChatGPT, Perplexity, Gemini or Google AI Overviews mention your brand at all, does it link to your site when it does, and does it actually recommend you over competitors. That's a different question from "do we rank on page one," and treating it as the same thing is the most common mistake teams make.
A traditional SEO audit grades ranking factors: title tags, backlinks, Core Web Vitals, keyword coverage. An AI visibility audit grades something adjacent but distinct, often called GEO (generative engine optimization) or AEO (answer engine optimization): whether a language model, when synthesizing an answer, pulls your content into its response and attributes it to you. A page can rank number three on Google and never get cited by an AI Overview, because the model picked its source from a different, more extractable page ranking seventh.
One more distinction matters: an audit is a snapshot, not a subscription. You run it once, record what you found, and move on. Continuous monitoring is the ongoing version of the same work, usually run through a platform that reruns your prompt list daily or weekly and tracks share of voice over time. Both have a place. The audit tells you where you stand today; monitoring tells you whether a fix actually moved the needle three weeks later.
How Do You Run an AI Visibility Audit in 5 Steps?
The fastest path to a usable audit is five steps run in sequence: pick prompts, run them across engines, log what comes back, trace the sources the model cited, then check why your own pages either got picked or got skipped. A basic pass across four engines and 10 to 15 prompts takes 45 to 60 minutes if you're doing it by hand with no tooling.
Step 1: Build a Buyer-Prompt List, Not a Keyword List
Keywords describe what people type into a search box; buyer prompts describe what people actually ask a chatbot, in full sentences, often mid-research. "best project management software" is a keyword. "What project management tool should a 15-person marketing agency use if we're already on Google Workspace?" is a buyer prompt. The second version is what someone actually pastes into ChatGPT, and it's the version that surfaces specific brand recommendations rather than a generic listicle structure.
Step 2: Run Prompts Across ChatGPT, Perplexity, Gemini, AI Overviews
Run the identical prompt list through each engine manually, or use a tracking tool to automate the repetition. Doing this by hand means opening ChatGPT, Perplexity, Gemini and running a Google search for each query to check for an AI Overview, then recording the raw output for all 10 to 15 prompts. At scale, this gets tedious fast. A prompt-tracking tool such as Peec AI automates the repetition and stores historical results, which matters once you're tracking the same prompt list monthly instead of once.
Step 3: Log Mentions, Position and Sentiment
For every prompt and every engine, record four data points: was the brand mentioned, where in the response, whether sentiment was positive, neutral or negative, and whether a link accompanied the mention. A brand named third in a five-brand list with a neutral tone is a very different result from a brand named first with an explicit recommendation and a working link. Treat sentiment analysis as part of the raw data, not an afterthought; a mention surrounded by hedging language ("some users report issues with...") scores worse than a flat factual mention even if both count as a "citation."
Step 4: Trace the Citation Source Graph
Every AI answer that cites a source is revealing part of its citation source graph, the set of domains it trusts enough to pull from and link to. Open every linked source in the response and note the domain: is it your site, a competitor's, a review aggregator like G2 or Capterra, a Reddit thread, a trade publication? Patterns emerge fast. If five of your ten prompts cite the same third-party roundup article, that page is functioning as a gatekeeper to AI visibility for your category, and you now know exactly who to pitch for inclusion or correction.
Step 5: Check Technical Extractability
Mentions without links usually trace back to a technical problem, not a content problem. Pull up the pages you'd expect an AI engine to cite and check whether they're actually parseable: clean HTML structure, no content locked behind JavaScript rendering that crawlers skip, fast load times, and a crawlable robots.txt. This step converts "we're not mentioned" into "we're not mentioned because the model can't read the page," which points to a fundamentally different fix.
How Do You Choose Buyer Prompts for an AI Visibility Audit?
Choose prompts that mirror the three stages of a real purchase path, not just head-term variants of your main keyword. Most audits fail here first, loading up on ten versions of "best [category] tool" and missing the questions that actually drive a buying decision.
| Stage | Example prompt | What it reveals |
|---|---|---|
| Awareness | "What's the difference between GEO and traditional SEO?" | Whether you show up in educational, top-of-funnel answers |
| Consideration | "What are the top AI visibility monitoring tools for a mid-size SaaS brand?" | Whether you're named in a direct comparison set |
| Decision | "Is [competitor] or [your brand] better for an agency managing 20 client accounts?" | Whether the model takes a position and whether it favors you |
| Objection-handling | "Does [your brand] support Google AI Overviews tracking?" | Whether the model has accurate, current feature data on you |
Aim for 10 to 15 prompts split roughly evenly across these four buckets. Pull real phrasing from sales call transcripts, support tickets or your own search console queries rather than inventing prompts in a vacuum. Generic prompts produce generic, low-signal results.
How Do You Score What You Find in an AI Visibility Audit?
Score every prompt-engine pair on a 0 to 3 scale, then average across the full set to get a single visibility score you can track over time. This is the step most audit guides skip entirely, leaving teams with a pile of screenshots and no way to judge whether the results are good, mediocre or alarming.
| Score | Label | What it looks like |
|---|---|---|
| 0 | Not mentioned | Brand absent from the response entirely, competitors named instead |
| 1 | Mentioned, no link | Brand named in passing, no citation or source link attached |
| 2 | Cited with link | Brand named with a direct link back to a page on your domain |
| 3 | Recommended with reasoning | Model explicitly favors the brand and explains why, not just lists it |
Sum the scores across all prompt-engine combinations and divide by the maximum possible score (3 times the number of combinations) to get a percentage visibility score. Take a 20-person marketing agency running 12 prompts across four engines, 48 combinations total, max possible score 144. If the raw sum comes to 58, that's a 40% visibility score, meaning the brand is mostly landing in the "mentioned, no link" tier with a handful of 0s and almost no 3s. That number alone tells you the priority isn't content volume, it's getting existing mentions upgraded to linked citations.
Segment the score by stage too. A brand scoring well on awareness prompts but 0 across every decision-stage prompt has a different problem than a brand invisible everywhere; the first needs sharper comparison content, the second needs a presence strategy from scratch.
Is Your Site Even Extractable?
Extractability is whether an AI model's crawler and parser can actually read, structure and reuse your content, and most sites fail this quietly without any ranking signal to warn them. A page can satisfy every traditional SEO check and still be invisible to a language model if it's JavaScript-rendered with no server-side HTML fallback.
| Signal | Why it matters | Quick check |
|---|---|---|
| Schema markup | Gives the model structured facts to lift directly | Validate with schema.org's reference docs |
| llms.txt | Signals which pages are meant for AI consumption | Check for a file at `/llms.txt`, per the llmstxt.org spec |
| Clean HTML / no JS-gating | Crawlers that don't execute JavaScript need server-rendered content | View page source, not just the rendered DOM |
| Crawlability | Blocked pages can't be cited, full stop | Check robots.txt and any noindex tags |
| Page speed | Slow pages time out during crawl passes | Run a PageSpeed Insights check |
| Structured FAQs | Q&A format maps directly onto how models answer questions | Add FAQ schema to high-intent pages |
Some monitoring tools now score crawlability as part of their platform rather than leaving it to a separate technical audit. Otterly.AI, for example, includes a crawlability scoring component alongside its prompt tracking, which saves a manual step if you're already running the platform for monitoring. Google documents how its AI features pull from indexed content in its AI features documentation, worth a read if you're trying to understand why AI Overviews specifically behave differently from a standard snippet.
DIY, Monitoring Tool or Agency Audit: Which Fits You?
The right choice depends on how often you need the data and how much budget you have for precision, not on which option is "best" in the abstract. All three answer the same underlying question; they differ in depth, repeatability and cost.
| Approach | Cost | Time to first result | Best for | Limitation |
|---|---|---|---|---|
| DIY manual audit | Free | 45-60 minutes | Small teams, first-time check, tight budget | No historical tracking, doesn't scale past ~15 prompts easily |
| Monitoring platform | Entry tiers from roughly $1 to $800/month, median around $79/month per hand-verified data from Geodeck's tool dataset | Same day | Ongoing tracking, teams that need to prove movement over time | Requires someone to act on the data, not just watch it |
| Agency-led audit | Roughly $500 to $3,000+ depending on scope | 1-2 weeks | Deep citation mapping, competitive analysis, remediation roadmap | Highest cost, slowest turnaround, one-time unless retained |
A DIY pass is the right starting point for almost everyone. It costs nothing but an hour, and it tells you whether you even have a visibility problem before you spend money finding out in more detail. If the DIY audit turns up a real gap and you need to prove progress to a stakeholder over months, that's when a monitoring platform earns its subscription fee; browse Geodeck's comparison of 38 AI visibility monitoring tools to see what fits your prompt volume and budget. If the findings are complex, span multiple brands or product lines, or you need a team to both run the citation-source analysis and build the remediation plan, that's an agency job; Geodeck's directory of GEO agencies lists firms that offer done-for-you audits with defined deliverables.
How We Evaluated These Options
This comparison is based on published pricing pages for monitoring platforms as of 2026, Geodeck's own hand-verified dataset of 64 GEO tools covering pricing tiers and feature coverage, and publicly listed scope-and-deliverable pages from agencies offering AI visibility audits. We did not run every tool ourselves for this piece; where a specific capability is attributed to a named product, it reflects that vendor's own documentation.
How Do You Turn Audit Results Into Fixes?
Prioritize fixes in the order they affect the most prompts, not the order they're easiest to implement. A single structural fix, like adding FAQ schema to your top ten comparison pages, often moves more prompt-engine combinations than ten individual content tweaks.
Work the backlog in three tiers. First, fix extractability problems on pages that already rank well in traditional search but scored 0 or 1 in your audit, since these are the cheapest wins: schema markup, llms.txt, server-rendered HTML. Second, close content gaps where your audit found zero coverage at the consideration or decision stage, usually meaning you need a direct comparison page or a page that explicitly answers the objection-handling prompts from Step 1. Third, invest in citation building, meaning outreach to the third-party sites your citation-source-graph trace in Step 4 revealed as gatekeepers; getting listed accurately on a G2 category page or a trade publication roundup often does more for AI visibility than another blog post on your own domain.
For the technical and content layer specifically, Geodeck's directory of GEO tools covers platforms built around schema generation, llms.txt setup and structured content formatting, useful if the audit surfaced extractability as the primary bottleneck rather than a content gap.
What Can't an AI Visibility Audit Tell You?
An audit can't guarantee future citations, and it can't promise the result you got today will repeat tomorrow. AI answers are non-deterministic by design: the same prompt run twice in the same session can return different phrasing, different source links, even a different brand mention, because the underlying model samples from a probability distribution rather than retrieving a fixed answer.
Results also vary by model version, and vendors update models without much notice. A brand cited reliably in March can drop out in April after a quiet model update, with nothing on the brand's side having changed at all. That's why a single audit is a snapshot of one moment, not a trend line; treat any one-off result with some skepticism, especially if it's unusually good or unusually bad compared to what you expected.
Session and account personalization add more noise. Logged-in ChatGPT sessions with memory enabled can return different results than a fresh incognito session, and geographic location shifts what Google AI Overviews and Gemini surface. None of this means audits are worthless, it means a single pass should inform direction, not serve as a precise scoreboard you report to the nearest percentage point.
FAQ
How long does an AI visibility audit take?
A basic manual pass across 10 to 15 prompts and four engines takes 30 to 60 minutes. A full agency audit with citation-source mapping and a remediation roadmap typically runs 1 to 2 weeks.
How much does an AI visibility audit cost?
Nothing, if you run it manually yourself. Monitoring tools start as low as a few dollars a month at the entry tier, with a median entry price around $79/month across published pricing. Agency-led audits generally range from $500 to $3,000 or more depending on scope and the number of prompts and markets covered.
How often should I re-run an AI visibility audit?
Quarterly, at minimum, for a manual snapshot audit. If you've moved to a monitoring platform, monthly or continuous tracking is standard, since AI answers shift whenever the underlying model updates, not on a predictable schedule you can plan around.
What's the difference between an AI visibility audit and a regular SEO audit?
A regular SEO audit evaluates search engine ranking factors: backlinks, on-page optimization, technical crawl health for traditional search indexing. An AI visibility audit evaluates something narrower, whether and how AI answer engines mention, cite and recommend the brand inside generated responses, which doesn't always correlate with search rankings at all.
Can I run an AI visibility audit for free?
Yes. Manually querying ChatGPT, Perplexity, Gemini and checking Google for AI Overviews costs nothing but time, roughly an hour for a first pass. Paid tools add automation, historical tracking and alerting on top, which matters once you're repeating the audit monthly rather than once.
What tools do agencies use to run AI visibility audits?
Agencies typically combine a prompt-tracking platform, such as Profound or Peec AI, with manual citation-source analysis and a technical crawlability check, sometimes using a platform like Otterly.AI that scores extractability directly rather than treating it as a separate manual pass.
Fact-checked against live sources, 2026-11-14: Verified that Peec AI (prompt/mention/position/sentiment tracking across ChatGPT, Perplexity, Gemini, AI Overviews), Otterly.AI (dedicated crawlability-scoring component), Profound (enterprise AI-visibility/prompt-tracking platform), and the llms.txt spec at llmstxt.org (Jeremy Howard, Sept. 2024, for signaling AI-relevant pages) are all accurately described; no corrections were needed to these tool/spec claims. Pricing figures, agency cost ranges, and dataset sizes attributed to Geodeck's own internal research could not be independently verified via external search and are presented as the publisher's own editorial data..