A citation in ChatGPT, Perplexity, or Google AI Overviews is a named mention of your company or product. It usually includes a link to a page the engine used as a source. This post walks through the AI visibility audit we run for B2B SaaS companies, then the loop we run after it. We watch citation share, draft a fix, and ship it under approval.
The implementation details already live elsewhere. The Next.js work is in our AEO/GEO guide. The Sanity defaults are in AEO/SEO best practices for Sanity. There's also a practical video introduction if you want the checks walked through on a real site.
What does "cited" mean in ChatGPT, Perplexity, and Google AI Overviews?
It means the answer names you, and often links a page it used. That's different from ranking a blue link. The engine can mention a competitor, mention you, mention both, or mention nobody in the category. That mix can change from one week to the next on the same prompt.
We treat AEO (answer engine optimization), GEO (generative engine optimization), and AI SEO as the same job. The service page says so directly: anyone selling them as three separate products is inventing scope.
Which questions should you track?
A prompt cluster is a group of questions that share the same job. "Which agencies offer an AI visibility audit for a B2B website?" and its close variants are one cluster. The audit looks at those clusters, not only at keywords the site already ranks for.
We don't publish a standard starter list in this post. The useful set is the questions your buyers already ask engines about the problem you sell. The competitor pass in the audit is there to show who is being named on those questions today.
Is this the same job as SEO?
Mostly yes. Crawlability, titles, metadata, structured data, and content quality still do most of the work. The diagnostic is what changes. Instead of only asking where a page ranks, you ask which prompts cite you, which engines do it, and whether share moved after a change.
The main technical addition is content negotiation. The same URL can serve HTML to a browser and clean markdown to an agent that asks for it. That pattern, plus JSON-LD, sitemaps, and llms.txt, is the subject of the Next.js AEO/GEO guide. If the site runs on Sanity, the Sanity AEO/SEO defaults cover how those fields and the JSON-LD should be derived rather than pasted in by hand.
You don't need a new CMS to start. The audit looks at the live site you already have.
What does an AI visibility audit check?
At Roboto Studio, an AI visibility audit is a fixed-scope review of your current citation footprint and technical readiness, written for a B2B or B2B SaaS website. It takes 1 to 2 weeks. You get a written report and a walkthrough call.
The checks are the ones already listed on the GEO service page:
- Content negotiation: whether key URLs can serve clean markdown when an agent requests it.
- Bot classification: whether you can tell real AI crawler traffic from spoofed user agents. Cloudflare's labeling is a common trap here. It files ChatGPT, Perplexity, and Claude under "AI Assistants", not "AI Search", so a rule that only targets "AI Search" misses all three.
- Schema: Service, FAQPage, BreadcrumbList, and Article where they help parsers, derived from the same content the page renders.
- llms.txt: a prioritized index rather than a flat list of every URL.
- Crawlability: list items in the DOM, navigation in static HTML, correct status codes, and a clean submitted sitemap.
- Competitor citation analysis across the prompt clusters you actually care about, plus a prioritized remediation roadmap ordered by expected impact.
We don't publish a typical "before" score or a guaranteed "after" score. Those figures aren't measured as a benchmark in this post. The audit's job is to say where you are cited, where you aren't, and which fixes to ship first.
Our service page cites Ramp's first-party experiment for the format point: markdown was the only format that reliably surfaced in LLM responses. That's why content negotiation sits at the top of the technical list. The wider Next.js setup, including how we do it on this site, is in the AEO/GEO guide.
Why publishing more pages often isn't the fix
A lot of AEO content is still taxonomy work: "What is GEO?", "GEO vs SEO". Models already have that, and it competes with Wikipedia and a long tail of near-identical posts. The pages that tend to get quoted are the ones with specific, evidence-backed claims that can't be lifted from a glossary.
Volume without editorial gates can move citation share the wrong way. The service page is blunt about this: LLMs weight source trust, so a corpus full of AI tells and unverified claims trains the wrong signal. The gates we put on a publishing pipeline are a slop check before merge, a fact-checker that tests claims against cited sources, and an editing pass before anything goes live.
What do you change after the audit?
The audit is a report. Setup is the implementation engagement, typically 4 to 6 weeks, scoped per site. On our service page that includes content negotiation on key pages, bot classification, llms.txt with prioritization logic, and schema where it's missing. It also includes restructuring high-value content so claims are specific enough to quote.
A monthly PDF of recommendations isn't that work. The gap we care about is the one between spotting a citation change and shipping the fix.
How fast share can move after a change isn't measured as a single figure. We say on the service page that a technical change or a content rewrite can move share within weeks rather than quarters. Speed still depends on your starting point and how competitive the prompt cluster is. That's as far as the evidence we're willing to put in this post goes.
How does the weekly loop run after setup?
A retainer keeps the same checks running on your stack. A background agent watches citation share on a weekly cadence. When share drops on a target prompt cluster, it flags what moved and drafts the fix. That fix might be a content edit, a schema patch, an llms.txt update, or a content negotiation rule. An engineer reviews it and ships it under your approval, then the agent watches that cluster again.
We documented how we built that agent for our own site in why we built our own background agent. For a client engagement, the same loop is wired to your data sources, your content repo, and your approval flow. The retainer on the service page has a minimum 3-month commitment, with a monthly synthesis call and a quarterly re-baseline.
The three engagement shapes, as published
These durations and deliverables are taken from our GEO service page, as of November 10, 2026. They aren't a quote, and they aren't a case-study result.
| Stage | What it covers | Duration |
|---|---|---|
| AI visibility audit | Citation footprint, technical readiness, competitor prompts, remediation roadmap, written report and walkthrough | 1 to 2 weeks |
| Setup | Content negotiation, bot classification, llms.txt, schema, content restructuring, monitoring dashboards | 4 to 6 weeks typical, scoped per engagement |
| Retainer | Weekly monitoring, agent-drafted fixes shipped under approval, monthly call, quarterly re-baseline | Minimum 3-month commitment |
What we will not claim
We don't publish a ranked list of "best AEO agencies for B2B SaaS", and we don't have a citation-lift number that we can honestly put on this page. If a guide promises a universal lift, ask for the sample, the date range, the engines, and the method.
If you want the checks walked through on a site, start with the practical introduction to AEO, GEO and AI SEO. The audit, setup, and retainer are on the AEO and GEO service page.
