Running AI Readiness Audits for Clients: An Agency Playbook

Clients are starting to ask about this whether or not an agency has a service built for it yet. "Are we showing up in ChatGPT?" "Should we be worried about AI search?" Those questions arrive whether or not there's a package ready to answer them, and answering "let me look into that" repeatedly is a worse position than having an actual, repeatable process. This is that process: how to check a site, present the findings, turn them into paid work, and keep it running, using the tools that already exist for it rather than assembling something from scratch.

Step one: check a prospect before they're a client

The best version of this workflow starts before a contract exists. The Chrome extension inspects any page entirely inside the browser, no login, no account, no crawl request sent anywhere, which means it works on a prospect's live site exactly the same way it works on a client's staging environment behind a password, or a page still sitting on localhost during a rebuild. That reach matters in a pitch conversation: you can open a prospect's own site during the call and show them something real about it, not describe a category of problem in the abstract.

What the popup actually shows, in order: a plain verdict for that specific page (clearly identified, partially identified, or unclear, with a plain-English reason why), the three most specific issues found, quoting the actual evidence rather than a generic description, and a comparison of what a normal browser sees against what an AI crawler sees when it requests the same page, which is exactly where cloaking, bot-blocking, or JavaScript-only content shows up. There's also a set of "ask an AI" buttons that open ChatGPT, Perplexity, Gemini, and Google's AI mode with a real question about the business pre-filled, so instead of asserting a problem exists, you can watch it happen live, in front of the client, on their own business.

Once you've found something worth raising, a one-click "copy diagnostic" turns that page's findings into ready-to-send text for an email or a message, which is the fastest path from "I noticed something" to an actual outreach note, without writing the explanation yourself from a blank page every time.

Worth being upfront with yourself about what this extension is and isn't, so the pitch stays accurate. It doesn't compute a full readiness score itself, that logic lives entirely on the server so the extension and a full audit can never quietly disagree with each other. If it shows an overall site score at all, it's the last real full-site audit's number, clearly dated, not a fresh calculation happening in the popup. Treat it as a fast, real, single-page verdict for a live conversation, and the full audit as the actual baseline you build a proposal on. The distinction between a single-page inspector and a whole-site check is the same one covered from the buyer's side in the three kinds of AI readiness checker, useful background if a prospect asks how this compares to something else they've already tried.

Step two: onboard the client and establish a baseline

Once there's a real engagement, add their site to your dashboard and run a full audit. This is the number everything else gets measured against, so it's worth treating this first run as a fixed reference point rather than something you'll casually re-run and compare loosely later. Multi-site accounts (five sites on the Premium plan, unlimited on Agency) are what make this practical once you're managing more than one client at a time, a single site switcher rather than separate logins per client.

Read the baseline pillar by pillar rather than leading with the single overall number in your own head. A client's first reaction to a low score is often defensiveness, "our website looks fine", and pillar by pillar is what turns a defensive reaction into a specific, believable conversation: not "your score is low" but "an AI system can't currently tell who runs your company," which is a concrete, checkable claim rather than a grade. This is also where a real content gap most often surfaces: plenty of established businesses have a website that looks complete to a human visitor and still says almost nothing explicit about who actually runs the company, a gap covered in more depth in why an AI can't answer basic questions about a real business, worth having open in a second tab the first few times you walk a client through their own findings.

If you're managing more than a handful of client sites, resist the urge to treat every account identically once they're all sitting in the same dashboard. A client on a slow six-month content cycle doesn't need the same monthly re-check cadence as one actively rebuilding their site right now. The site switcher makes it easy to jump between accounts; it doesn't make every account the same shape of engagement, and standardising your own internal process too rigidly across a varied client roster tends to produce busywork on the quiet accounts and not enough attention on the active ones.

Step three: present findings as a client actually sees them

This is where the difference between a freelancer with a login and an agency with a service shows up most. On the Agency plan, reports can go out under your own brand: your logo, your company name, your accent color, your own contact email, and a custom line of your own choosing, replacing the defaults entirely.

One thing stays fixed regardless of plan, worth knowing before you promise anything to a client: a small, permanent line crediting the AI Readiness score to AI Schema Gen. It can't be turned off, and it's a deliberate choice, not an oversight: the score is free for any domain specifically so it can be cited by people who never pay for it, and a completely anonymous number would undercut the reason it's free in the first place. In practice this reads the way a "results calculated with [tool]" line reads on plenty of professional reports, present, small, and not a barrier to putting your own name and relationship front and center on the rest of it.

Step three and a half: get the site actually connected, not just audited

This is the step that catches people who assume a completed audit means they're ready to fix things. It doesn't, and it's worth understanding why before you scope a project timeline around it.

A site claimed through a free audit starts with no write access at all, an audit only ever reads a site from the outside, it never requires anything be installed. Every one-click fix, by contrast, needs a real, verified connection back to that specific site: on WordPress, that means the plugin is actually installed and has synced at least once; on a Next.js site, it means the @aischemagen/nextjs package is deployed and has proven it's actually live, not just that a token was generated and setup was started but never finished. Until one of those is true, the fix buttons simply don't appear, on purpose, because generating a fix that can never reach the live site would still spend the client's monthly quota for a result nobody would ever see.

In practice, this means the connection step is its own task in a client engagement, usually meaning getting a developer or the client's own team to install a plugin or ship a small deploy, and it's worth putting a specific line for it in your project timeline rather than assuming it happens automatically once someone signs off on the audit findings. A baseline audit can be run and presented the same day a contract is signed; the fix work genuinely cannot start until this separate step is done.

Step four: turn findings into paid work, not just a report

A findings list by itself is a diagnosis, not a deliverable. The actual paid work is closing the gaps, and it splits cleanly into three kinds of effort worth pricing differently, because they cost you different amounts of time. Some findings resolve with a single click, no manual work on your side at all. Others need one round of questions answered by the client (their founder's name, their service area, their registration details) which you then apply for them. The rest is genuine content or engineering work, missing pages, thin descriptions, technical fixes, real billable hours. The exact breakdown of which is which, and why some technically simple problems still don't have a one-click fix, is covered in which AI readiness problems are machine-fixable, and it's worth reading before you scope a client engagement so you're not promising "we'll automate everything" when a real third of the list is genuinely manual.

Step five: make it a recurring service, not a one-off report

A single audit is a good opener; a recurring check is the actual retainer. Re-run the audit on a regular cadence (monthly is reasonable for most clients) and show the client the trend, not just the current snapshot, movement over time is what justifies an ongoing fee far better than a single static score ever will.

Competitor benchmarking is a natural add-on once a client's own baseline is established: tracking one or two of their real competitors and showing where the client leads and where they're behind gives you a second, genuinely different angle to check and report on. The mechanics of running that comparison well, and reading the results without drawing the wrong conclusion, are covered in how to benchmark AI readiness against a competitor. Competitor checks are metered per month by plan, so build your retainer cadence around what your plan actually allows rather than promising unlimited comparisons up front.

This whole sequence, audit, fix, benchmark, re-check, is the same loop covered end to end in the AI readiness loop, written for a single site owner rather than an agency managing several, but the underlying stages are identical, and it's a useful shared reference to point a client at if they want to understand the shape of the ongoing work themselves.

Be honest with clients about where this practice is still developing. Readiness (can an AI find, read, and correctly describe the business) is measurable and actionable today. Whether a specific AI assistant is actively citing the business in live conversations is a separate, harder-to-observe question that the industry, including this product, is still building reliable tooling for. Promise the part that's solid; don't oversell the part that isn't yet.

Thinking about what to charge

This isn't going to hand you a specific price, that depends on your market, your positioning, and what else you bundle it with, and any number offered here without knowing your business would be a guess dressed up as advice. What's worth doing instead is pricing from a real cost base rather than a feeling. The Agency plan itself is a fixed, known monthly cost, so start there: work out how many client audits, fixes, and re-checks that one cost has to cover before the service line is profitable at whatever margin you're aiming for, then price the service around that math rather than around what feels like a round number. A retainer built to comfortably clear its own tool cost, with real margin on top for the actual fix work (which is where the genuine hours go, not the audit itself), holds up better than a flat rate picked before you've run a single real engagement.

Working with a team, honestly

Team seats (three on Premium, unlimited on Agency) let you bring colleagues into the account to help manage client work. Worth knowing the actual shape of this before you plan around it: a seat gives someone editor or viewer access to the whole account, every connected site, not a scoped view of just one client's site. That makes it a good fit for your own internal team dividing up client work across a shared account. It's not a mechanism for giving an individual client their own login limited to seeing only their own results, if a client wants that kind of visibility, sharing the branded report link or PDF directly, rather than an account seat, is the more accurate fit for what a seat actually does today.

Common mistakes

Leading a pitch with the score instead of a specific finding. A number means little to someone who's never seen the scale before. "An AI couldn't answer a basic question about what you do" lands harder than "you scored 54."

Re-running the baseline audit and treating the new number as directly comparable without checking what changed. A site redesign, a new page structure, or even a crawl catching a temporary error can move a number without any real readiness change. Read what actually moved before reporting a trend.

Scoping a fix timeline before the site is actually connected. An audit can be run and presented on day one; the fix buttons don't exist until the site has a real, verified write connection back to it. Get the plugin installed or the package deployed early, not as an afterthought once the client is already expecting fixes to start.

Promising full automation on the entire findings list. A genuine share of most sites' problems needs real content work or an engineering fix, not a button. Set that expectation at the scoping stage, not after the client assumes everything was one click.

Spending the monthly competitor-check quota on curiosity rather than a plan. Decide which one or two competitors are actually worth tracking for a given client before you start clicking through prospects at random.

Turning on branding without checking it end to end first. Branding is off by default even after you configure it, a deliberate safeguard so a half-finished setup never accidentally goes out under your name. Send yourself a test report before the first client-facing one goes out.

Treating every finding as equally worth fixing first. A findings list sorted by severity alone isn't the same as one sorted by actual value. Some fixes recover a meaningful chunk of score for very little effort, others cost real hours for a small gain, and scoping a client engagement around the first list without checking which is which leads to either overcharging for easy wins or underpricing genuine work.