The first AI search audit and proposal: what agencies should check before the meeting and what they can promise
Build the first AI search optimization proposal on three things: audit the site before the meeting and show the specific facts AI cannot read; commit to the work, the schedule, how often you report and how you measure; and do not promise mentions in AI answers or positions. In June 2026 Google revised its guide to hiring an SEO and now suggests that site owners check whether AEO and GEO advice follows Google's official guidance. Write the proposal expecting it to be read that way.
Written November 15, 2026 (JST). This is for web agencies, SEO firms, local SEO firms and WordPress care companies pitching AI search optimization to a client (a shop, a clinic, a small company) for the first time. How to run it as a monthly service is covered in offering AEO as a monthly service; this guide stays with the step before that: the first audit and the first proposal.
AI answers change from day to day and with the wording of the question. The steps here are for telling clients accurately where they stand and what the work covers. They do not guarantee that a business will appear in AI answers.
The 2026 baseline: clients can check your pitch against Google's guide
Google's "Do you need an SEO?" page (last updated June 5, 2026) now lists "Optimizing for generative AI" among the services SEOs offer, and has a new section on evaluating an SEO's recommendations and tools. It suggests checking whether advice about AI experiences (AEO or GEO) matches Google's guidance on optimizing for generative AI. The same page says: "No one can guarantee a #1 ranking on Google" (Google Search Central: Do you need an SEO?).
A companion page, "Google Search's guidance on using third-party SEO tools, services, and advice", names AEO and GEO tools among third-party services and says Google doesn't evaluate third-party services, that any predictions such tools make may not happen, and that using a tool doesn't guarantee success. Good advice, it says, either labels its claims as opinion based on data or experience, or cites official Google guidance (Google: third-party SEO tools, services, and advice).
In a proposal, that comes down to three rules.
- Keep what Google and other vendors document apart from your own experience and opinion, and link the official sources.
- Do not write "Google-approved", "guaranteed to appear in AI answers" or "will rank higher".
- Say what results you expect, by when, and how you will measure them. Google's guide suggests owners ask about results and timeframe, how success is measured, and whether all site changes will be shared (same page); answer those in the proposal before you are asked.
The pre-meeting audit: what to check and how to write it up
Using public information only, a first audit takes about 30 to 60 minutes. The goal is not a score. It is to show the facts AI cannot read from the site, in a way the client can verify with their own eyes.
| What to check | Where to look | How to word it in the proposal |
|---|---|---|
| Indexing and snippets | URL Inspection in Search Console (or a site: search without access); noindex and nosnippet | "The home page and pricing page are eligible for Google's AI features" |
| AI search crawlers | Whether /robots.txt blocks OAI-SearchBot, ChatGPT-User, PerplexityBot, Perplexity-User and similar | "ChatGPT's search crawler is blocked in robots.txt" |
| Business structured data | The page's JSON-LD: does the Organization or LocalBusiness carry the address, phone and opening hours? | "The structured data names the company and logo but has no address or opening hours" |
| /llms.txt | Is it at the root, and does it point to current pages? | State the fact only: "There is no llms.txt" |
| Conflicting facts | Hours, address and prices on the site, the Google Business Profile and directory sites | "The site says closed Wednesdays; the directory says Tuesdays" |
| Text AI can read | Are prices, hours or menus only in images? | "The price list is an image and cannot be read as text" |
| AI answers (a sample) | A few questions with the business name, and a few with only the business type and area | "Results of these three questions, asked twice each today (answers vary)" |
OpenAI and Perplexity publish their crawler names and purposes (OpenAI: overview of crawlers, Perplexity: crawlers). Google says a page needs to be indexed and eligible to be shown with a snippet to appear in its AI features, and that SEO best practices apply to them as usual (Google: AI features and your website). With access to the client's Search Console you can also read impressions in Google's AI features in the generative AI performance report, available to all sites as of August 31, 2026 (Search Console Help).
Three cautions when you use AI answers in an audit
A single run tells you little. In a study by AI検索ドック (Stellagent) in October 2026 (30 questions, 3 AIs, 10 runs each), one run differed from the 10-run rate by 29.9 points on average, three runs by 16.5 and five runs by 10.7 (AI検索ドック, a study by a tool vendor). In the first meeting, avoid saying "you don't appear"; present it as "the result under these conditions today".
- Separate questions that include the business name from discovery questions with only the type and area. A named question will usually return the name; what matters there is whether the facts are right.
- Record the date, the AI, the number of runs and whether you were signed in, so you can repeat it under the same conditions next month.
- If a competitor shows up, do not use it against them. State the fact: "for this question, this company was cited".
There is research that AI summaries reduce clicks on regular results. Pew Research Center looked at the March 2025 browsing of 900 U.S. adults: when an AI summary appeared, a traditional result was clicked in 8% of visits, against 15% when none appeared, and a link inside the summary in 1% (Pew Research Center). It is a U.S. study; say that it may not carry over to another country or a particular site.
The proposal in seven parts
A first proposal does not need to be long. Fit it into seven parts so the client can pass it around internally and decide.
| Part | What goes in | What stays out |
|---|---|---|
| 1. Summary | The top three problems found and a one-line description of the work | A preamble of industry generalities |
| 2. Where the site stands (audit) | The results of the table above, with dates, methods and screenshots | A proprietary score with no explanation |
| 3. Why fix it | Questions the client is likely to be asked, matched to facts AI cannot read | Unsourced market figures, or a study's uplift presented as the client's forecast |
| 4. Scope of work | What will be fixed, who does each item (agency or client), the list of changes to the site | "Full AI optimization package" with nothing behind it |
| 5. Schedule | Week-by-week plan for the setup, and the dates the client reviews | "Appear in AI within X weeks" |
| 6. Reporting and measurement | How many questions you track, how many runs, which AIs, what the report covers and when | Treating one measurement as a result |
| 7. Fees and terms | Setup fee and monthly fee, what is included, how to cancel, who pays for tools | Pay-for-ranking guarantees |
We found no primary source for what agencies charge for monthly AI search work, so this guide gives no market price. For audits alone, Stellagent compared 24 Japanese diagnosis services and found paid entry prices from 10,000 to 900,000 JPY, most often 100,000 to 165,000 JPY (checked on each vendor's page on October 7, 2026; AI検索ドック, a company that also sells audits).
What you can promise, and what you cannot
| You can promise | You cannot promise |
|---|---|
| Finishing the agreed work by the agreed date | Appearing in ChatGPT's or Google's AI answers |
| Measuring under the same conditions every month and reporting on a set date | A position in AI answers or an X% increase |
| Sharing every change made to the site | A method approved by Google or certified by an AI company |
| Separating what official documentation says from your own experience | That an uplift shown in a study will happen on the client's site |
| Investigating and reporting the cause if the numbers drop | Bringing a dropped number back |
How to explain a drop is covered in why AI visibility dropped. Writing "if the numbers drop, we investigate in this order and report" into part 6 of the first proposal makes that conversation easier later.
Common client questions and how to answer them
| The client asks | A way to answer |
|---|---|
| Is this separate from SEO? | Much of the work overlaps; Google says SEO best practices apply to its AI features. What is added is organizing the facts AI reads (structured data, llms.txt, crawler access) and measuring how AI shows the business. The differences |
| When will it work? | We cannot promise a timeframe. We commit to when the setup is finished and from when we measure under the same conditions each month. |
| How will we know we appear in AI? | Each month we ask a fixed set of questions a fixed number of times and record whether the name appeared and the site was cited. For Google, Search Console's generative AI performance report also shows impressions. |
| Can we keep our SEO plugin? | Usually, yes. We check what it already publishes first, so the business is not described twice in structured data. |
| Will adding llms.txt get us cited? | We cannot say it will on its own. It is one of the files AI can read, and we add it alongside the rest of the work. Writing llms.txt |
Running the audit and setup on WordPress (with Nurevo)
If the client's site runs WordPress, the free Nurevo AEO plugin covers much of the audit and setup from the admin screen. Here is the order to work in as an agency.
- Before the meeting: put the client's URL into the free check, which needs no sign-up, and see the state of structured data, llms.txt and AI crawler access. Use the result screen in part 2 of the proposal.
- After you win the work, before installing: tell the client you will install it on their site (the partner program page has a notice you can send as is).
- Install: add the plugin from WordPress.org and go through the setup wizard (business type, basic facts, what to publish for AI, diagnosis). Automatically extracted details only fill blank fields, so check and correct them.
- Diagnose: on the AEO Score screen, read what the diagnosis sends (the site URL and language), tick the consent box and press Run diagnosis. You get a 0 to 100 score and a list of fixes; match it against the scope in part 4.
- Set up: llms.txt, basic structured data and access for AI search crawlers can be fixed right there. On sites with Yoast SEO, Rank Math, All in One SEO or Slim SEO, only the missing facts are added to that plugin's structured data, so nothing is published twice.
- Start measuring: once the site is connected, Free and Standard sites get a monthly check that asks Perplexity three questions built from the business type and area (never the business name) and shows whether the business was named, the site cited, and who appeared instead. Use it as the first record for part 6.
- Roll out to other clients: export the settings as JSON under Settings > Tools or with WP-CLI (wp nurevo export / import) and import them elsewhere. Keys, tokens and the plan are never exported.
For agencies there are monthly client reports under your name and logo sent automatically, a dashboard for several client sites, and Standard plan payment links to send clients. Paid plans are taken out by the client, with the client's own consent. A referral commission is calculated monthly: 40% of each referred client's payment net of Stripe fees (from the second referred client; approval required; overseas partners may also apply. For overseas payouts, Wise country, currency and recipient eligibility is reviewed individually; payment method, timing and fees are communicated in advance. Automated payouts in Japan use Stripe Connect). See the partner program.
To be fair, dedicated measurement tools track more AIs and more questions, and Pro's AI visibility measurement currently covers Perplexity (beta). Pairing Nurevo with a measurement tool is a reasonable choice (AEO tools compared).
Checklist before you send the proposal
- The audit date, method and screenshots are kept.
- Indexing, robots.txt, structured data, llms.txt, conflicting facts and image-only information were checked.
- AI questions are split into named and discovery, with date, AI and number of runs recorded.
- Official facts have sources; your opinions are labeled as opinions.
- The scope is written with page names and URLs.
- The monthly measurement conditions (questions, runs, AIs) and the reporting date are written down.
- No "guaranteed", "top position" or "Google-approved" wording is left.
- It says you will tell the client before installing anything, and that paid plans are taken out with the client's consent.
FAQ
Should the first audit be free?
There is no rule. One split is to make the public-information audit (the table in this guide) a free sales document, and charge for a deeper audit with Search Console access or competitor research. Either way, writing down what you checked makes your audit easier to compare.
Can we install a plugin on the client's site to audit it?
Before you win the work, auditing from public information only is the safe choice. Install anything only after the client agrees. Nurevo's free check (/check) needs only the URL.
Can the proposal mention that a competitor appears in AI answers?
Yes, as a fact with the date, question and AI. Answers change, so instead of "you are losing to them", write "under these conditions, this company was cited". Avoid running the competitor down.
AEO, GEO, LLMO or AIO: which term should the proposal use?
Pick the one the client has heard of and pair it once at the start, for example "AI search optimization (also called AEO or GEO)". Google's guide refers to "AI experiences (sometimes called AEO or GEO)".
Can we offer a pay-for-results contract?
Because AI answers vary for the same question and no one can guarantee a mention or position, we advise against making AI appearances the condition for payment. Contract for completed work and for measuring and reporting under agreed conditions.
Does this work for sites not built on WordPress?
Yes. The audit items (indexing, robots.txt, structured data, llms.txt, conflicting facts) are the same on any site. Nurevo covers non-WordPress sites with a snippet and businesses without a website with a hosted page.
If we pitch this way, will the client appear in AI answers?
That cannot be guaranteed; AI answers change by day and wording. These steps are for telling the client accurately where they stand and what the work covers, so they can compare and decide.