---
title: "The AI Search Recovery Proposal I Wouldn’t Sign"
description: "A line-by-line teardown of a typical AI search recovery proposal: what to cut, what to keep, and how to judge the work by enquiries instead of traffic and mentions."
url: "https://groas.com/post/we-took-apart-a-typical-ai-search-recovery-agency-proposal"
image: "https://pub-87da24ecbbfc4c3bad6875f3aa013712.r2.dev/generated-images/b08c60ee-1012-4265-9ace-d5bb3f0f8d5c.png"
published: "2026-11-02T06:04:05.623Z"
modified: "2026-11-02T06:04:05.677Z"
---

November 2, 2026 · 10 min read

# The AI Search Recovery Proposal I Wouldn’t Sign

[Alexander PerelmanHead Of Product @ groas](https://groas.com/author/alexander-perelman)[LinkedIn](https://www.linkedin.com/in/alexander-433793253/)

![An open agency proposal with nearly every line struck through in red, its signature line left blank, a capped pen set aside, and a silent desk phone behind.](https://pub-87da24ecbbfc4c3bad6875f3aa013712.r2.dev/generated-images/b08c60ee-1012-4265-9ace-d5bb3f0f8d5c.png)

In this article

1. [1. The traffic-recovery promise starts with the wrong target](#1-the-traffic-recovery-promise-starts-with-the-wrong-target)
2. [2. The 30-page refresh follows the losses, not the buyers](#2-the-30-page-refresh-follows-the-losses-not-the-buyers)
3. [3. The GEO package mixes useful work with mention theatre](#3-the-geo-package-mixes-useful-work-with-mention-theatre)
4. [4. The two-team org chart leaves the buyer between them](#4-the-two-team-org-chart-leaves-the-buyer-between-them)
5. [5. The monthly report counts everything except the decision](#5-the-monthly-report-counts-everything-except-the-decision)
6. [6. The fee page asks you to buy the add-on before proving the work](#6-the-fee-page-asks-you-to-buy-the-add-on-before-proving-the-work)
7. [The four-page proposal I would keep](#the-four-page-proposal-i-would-keep)
8. [Seven questions to ask before you sign](#seven-questions-to-ask-before-you-sign)

A twelve-page _AI Search Visibility Recovery Plan_ promises to win back organic traffic, refresh 30 informational pages and add a GEO package. I would not sign it. I chose this composite proposal because its line items show the problem with many AI search recovery pitches: they sell lost visits and AI mentions back to you without showing how either becomes an enquiry.

We already put the click-versus-citation question on the table in [_AI Overviews Took the Click. Should You Fight for It or the Citation?_](https://groas.com/post/lost-traffic-to-ai-overviews-fight-for-the-click-or-fight). This is the second look: once you accept that AI Overviews will keep some clicks, who should you pay, and for what? I am taking this proposal apart in the order a buyer would read it. Each line gets a verdict. **The work worth keeping must connect to enquiries across Google and AI assistants.**

## 1. The traffic-recovery promise starts with the wrong target

Page one opens with a chart of falling clicks and a promise to build them back. The chart may be accurate. The promise ignores why the clicks fell. In a randomized test of AI Overviews, showing the box [cut outbound organic clicks 39.8%](https://ppc.land/researchers-find-google-ai-overviews-cut-publisher-clicks-39-8/) and pushed zero-click searches up 34.5%. In a larger analysis, Ahrefs found the [top-ranking page’s click-through rate was 58% lower](https://ahrefs.com/blog/ai-overviews-reduce-clicks-update/) when an Overview was present.

The mechanism matters more than either percentage. The answer now appears above the old organic results. Move a page from position three to position two and you may still be asking a reader to click after Google has answered the question. Our own search data includes a query at position three with roughly 4,000 impressions and one click; the screenshot and its implications are in [_Position 3, 4,033 Impressions, 1 Click: What Our Search Data Says About SEO Traffic_](https://groas.com/post/4-033-impressions-1-click-at-position-3-what-our-own-search). A better ranking is not a repair for an answer the searcher no longer needs to leave Google to get.

**Verdict: cut the promise to restore the old session count.** Ask instead which pages still give a potential customer a reason to visit, and how the agency will count the enquiries those pages produce. That is a narrower promise, but at least it addresses the cause.

## 2. The 30-page refresh follows the losses, not the buyers

Next comes a list of the biggest traffic losers, followed by a plan to rewrite them. Think _what is a heat pump_, _how much does solar cost in 2024_ or _types of commercial insurance_. These explainers once brought in visitors. That does not make every lost visit a lost customer.

Ahrefs examined 146 million searches and found [99.9% of keywords that trigger AI Overviews are informational](https://ahrefs.com/blog/ai-overview-triggers/), while 5.5% are commercial and 1.2% are transactional. Those categories can overlap, but the direction is clear: the affected searches lean heavily toward questions an answer box can settle. A sharper introduction to an explainer will not necessarily give that searcher a reason to click through.

I used to tell clients to defend those pages first because they carried the site. I was wrong to treat the traffic total as the priority. A buyer asking for _emergency boiler repair near me_ or a _commercial groundworks quote Bristol_ still needs a vendor. An answer box cannot send the van. The proposal should start with pages that help that buyer decide whether to call, not with a spreadsheet sorted by lost sessions.

**Verdict: keep a refresh only where the page serves a buyer question and can be tied to an enquiry.** If the agency cannot distinguish those pages from explainers that mainly drew visits, strike the 30-page target. I would rather see a defensible list of ten buyer pages than thirty rewrites justified by yesterday’s traffic.

## 3. The GEO package mixes useful work with mention theatre

The next line sells a GEO package for ChatGPT, Perplexity and Gemini: AI schema, an `llms.txt` file, prompt monitoring and a monthly mention report. It sounds technical. Some of the underlying work is useful. But [Google’s eligibility guidance, summarized here](https://pages.citepoint.ai/articles/geo-for-agencies), does not require special AI schema or an AI text file. A page needs to be indexed and eligible to appear with a snippet.

For an assistant, the practical questions are less theatrical. Can it access the page? Does the page give a specific, current answer? Is the offer consistent with what a reader finds elsewhere? A dated changelog with a concrete update gives a system something to retrieve and quote; a vague service page gives it much less. Even a successful fetch proves only that a page was retrievable for that prompt at that time. It does not prove that the brand will appear in every future answer.

Then there is the source the assistant chooses. In a Derivatex test of 40 B2B SaaS categories, [only 11.6% of ChatGPT recommendations cited the tool’s own website](https://kairosy.ai/blog/why-does-chatgpt-cite-competitors-instead-of-your-brand); 88.4% pointed to a third party such as a review category page, Reddit thread or editorial roundup. A brand cannot solve that solely by adding another paragraph to its own site. It has to look at the lists and reviews appearing in answers for its category, then decide where its information or evidence is missing.

The useful jobs are plain enough to name:

- Check that relevant crawlers can access the commercial pages.
- Keep pricing, availability and dated proof current where those details appear.
- Make answers specific enough to be retrieved and understood.
- Identify the third-party pages assistants already draw on and where the brand is absent.

**Verdict: keep the access, content and third-party work; cut the guaranteed mention count.** Answers change between runs, with [fewer than one in 100 chances of ChatGPT giving the same brand list twice](https://pages.citepoint.ai/articles/geo-for-agencies). I would not pay a premium for a promise that depends on a list staying still.

## 4. The two-team org chart leaves the buyer between them

Halfway through, the proposal introduces an SEO team for Google and a GEO team for assistants. That looks like thorough coverage until both teams need the same commercial page to be accurate.

Take a plumber in Leeds. The service page says emergency call-outs within 90 minutes and carries last month’s prices and three named reviews. Google needs an indexed page that can appear in search. An assistant needs a page it can access, along with consistent information on the review and list pages it uses. These are different retrieval paths to much of the same buyer-facing information. If one team rewrites the page while the other pursues a third-party mention, who checks that the price and offer still match?

The answer cannot be two reports stapled together at the end of the month. Someone has to own the commercial page, its supporting evidence elsewhere, and the path from discovery to enquiry. That person should be able to tell you what changed, where it changed and whether calls or forms followed. A separate target for Google rankings and another for AI mentions makes that harder, not easier.

**Verdict: keep the work only if one owner is accountable for enquiries from both paths.** If the proposal has separate teams, separate success measures and no shared enquiry number, the org chart is overhead dressed as coverage.

## 5. The monthly report counts everything except the decision

Now the proposal shows its reporting sample: rankings, sessions and an AI visibility score based on brand mentions. These figures can help diagnose what happened. They cannot, on their own, tell a heating company whether more customers called.

A [reporting guide calls impressions, clicks, average position and keyword counts diagnostic rather than business outcomes](https://seojuice.com/blog/white-label-seo-reports/). That distinction should govern the order of the report. Put twenty rising terms and three AI mentions on page one, and you invite the client to mistake movement for progress. Put enquiries by landing page there, and the rest of the figures have a job: helping explain why the enquiry number changed.

The usual traffic chart has another limit. Search Console groups AI Overview clicks under the normal Web search type, so native reporting does not neatly separate them from ordinary clicks. A mention count has a different problem: it says nothing about whether the prompt came from a buyer. The same reporting guide argues for [qualified leads, calls, sales or attributable revenue first, followed by traffic split by landing page and intent](https://seojuice.com/blog/white-label-seo-reports/). That is the right order of attention.

I would want page one to show enquiries by landing page, with commercial and informational pages distinguished. Rankings, branded searches and citations can sit underneath as supporting evidence. If tracking cannot connect every assistant exposure to a call, say so; do not fill the gap with a visibility score and call it revenue.

**Verdict: keep the report if enquiries lead it.** If page one is rankings and mentions, strike the reporting template before it becomes the definition of success.

## 6. The fee page asks you to buy the add-on before proving the work

The last page makes GEO a separate retainer with a six-month minimum. That is not an implausible shape for this market: a published rundown lists [The HOTH AI Discover at $1,500–$10,000 a month, WebFX AI SEO from $3,000 a month, and uSERP from $10,000 a month on a six-month term](https://pages.citepoint.ai/articles/geo-for-agencies). The question is not whether an agency can charge a retainer. It is what the retainer commits it to doing, and how soon you can judge that work against enquiries.

A long term attached to a vague traffic-recovery promise puts the risk on the client. So does a separate GEO charge whose deliverable is a mention report. The agency can complete every listed task while the commercial pages produce no additional calls. That is the incentive problem, not an objection to paying for competent work.

I made the parallel case for paid search in [_7 Google Ads Agency Pricing Red Flags To Avoid In 2026_](https://groas.com/post/google-ads-agency-pricing-red-flags-2026-hidden-fees-to-avoid): check what triggers a higher fee, what a setup charge buys you and how easily you can leave. For reference, groas prices SEO and AI search at a flat $1,499 a month per service, month to month with no setup fees. A flat fee alone does not produce an enquiry. It does remove a reason to keep selling extra hours or an indefinite add-on as the answer.

**Verdict: strike a separate retainer or long lock-in when the enquiry target is vague.** Keep a clear, cancellable fee attached to named work and reporting on calls, forms and booked jobs by landing page. You should be able to tell what you bought before the first invoice arrives.

## The four-page proposal I would keep

After that teardown, I would rewrite the twelve pages as four. The shorter document would still have work to do; it would just stop treating every measurable activity as a result.

1. **Page one: establish the enquiry baseline.** Name the commercial and local landing pages, show which calls and forms can be tied to them today, and identify what the current reporting cannot tell us.
2. **Page two: choose the buyer questions.** List the pages worth improving because a potential customer uses them before acting. Leave the rest alone unless there is a separate reason to work on them.
3. **Page three: assign the access and trust jobs.** Check crawler access, current offers, dated answers and proof. Name the relevant third-party lists and review pages, and give one owner the work across Google and assistants.
4. **Page four: show the report before the work starts.** Put **enquiries by landing page at the top**, with citations, rankings and traffic underneath as evidence, not substitutes.

That version cuts the 30-page refresh justified only by lost visits. It cuts the separate GEO package built around schema and mention counts. It also cuts the six-month commitment before anyone can show what the work is doing. None of those cuts means ignoring AI assistants or abandoning informational content. It means giving each proposed task a reason to exist beyond filling a line in a scope of work.

## Seven questions to ask before you sign

1. Which landing pages will you tie to enquiries in my CRM, and what is the baseline today?
2. Which losing pages serve buyers, and which mainly brought visits you cannot win back?
3. What will you change so Google and assistants can access and use the relevant pages?
4. Which third-party lists and review pages appear in answers for my category, and where am I missing?
5. What appears on page one of the monthly report: enquiries by landing page, or rankings and mention counts?
6. Who owns the Google and AI search work against the same enquiry number?
7. What is the monthly fee, is there a setup charge, and can I leave after one month if enquiries do not move?

**Overall verdict: I would reject the recovery promise and keep the buyer-page work, the access and trust fixes, one accountable owner, and enquiry-first reporting.** If an agency cannot answer those seven questions plainly, it is selling the old traffic back to you. I would keep the pages that help a buyer act, make them usable across Google and assistants, and count what happens next. That is the filter I would apply to any agency, and the standard groas holds itself to: one owner, a flat fee, and work judged by calls, forms and booked jobs.

## Frequently Asked Questions

### Why shouldn’t an agency promise to restore the site’s old organic traffic?

Because AI Overviews can answer a search before the reader visits a website, restoring old rankings or session counts may not restore the lost clicks. The proposal should identify which pages still give potential customers a reason to visit and measure the enquiries those pages produce.

### Which lost pages should an agency prioritize for a refresh?

Prioritize pages that help potential customers decide whether to contact or hire the business, and that can be tied to an enquiry. A page that mainly drew informational visits is not automatically worth rewriting just because it lost traffic; a defensible list of ten buyer pages can be more useful than thirty traffic-led rewrites.

### What GEO work is worth keeping in an AI search proposal?

Keep work that checks crawler access to relevant commercial pages, keeps offers and dated proof current, makes answers specific, and identifies third-party lists and review pages assistants use. The proposal argues against paying for special AI schema, an \`llms.txt\` file, or a guaranteed mention count as the core deliverable.

### Who should be responsible for SEO and AI search work?

One owner should be accountable for the commercial pages, supporting evidence elsewhere, and the path from discovery to enquiry across Google and AI assistants. Separate teams and reports are not enough unless they share responsibility for the same enquiry number and someone checks that the buyer-facing information stays consistent.

### What should an agency’s monthly search report show first?

It should show enquiries by landing page first, distinguishing commercial and informational pages. Rankings, traffic, branded searches, and citations can help explain changes, but they are supporting evidence rather than business outcomes. Search Console groups AI Overview clicks under the normal Web search type, so it does not neatly separate them.

### When should a business reject a separate GEO retainer or long-term commitment?

Reject it when the enquiry target is vague or the deliverable is mainly a mention report. Keep a clear, cancellable fee attached to named work and reporting on calls, forms, and booked jobs by landing page, so the business can tell what it is buying before the first invoice.

## Paid Ads & SEO, Run For You

Businesses pay one flat monthly fee per service, agencies resell it under their own brand, and groas does the work either way.

[See Pricing](https://groas.com/pricing)

## Structured data

```json
{"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://graph.groas.com/entity/groas","name":"groas","url":"https://groas.com/","description":"groas is a done-for-you marketing service, not software: elite people working with proprietary technology systems run search ads, Meta ads, SEO and AI search for businesses, and white label for agencies, with a dedicated senior strategist owning every account.","logo":{"@type":"ImageObject","url":"https://groas.com/icon-512.png","width":512,"height":512},"sameAs":["https://www.linkedin.com/company/groas/"]},{"@type":"WebSite","@id":"https://groas.com/#website","name":"groas","url":"https://groas.com/","inLanguage":"en-US","publisher":{"@id":"https://graph.groas.com/entity/groas"}},{"@type":"WebPage","@id":"https://groas.com/post/we-took-apart-a-typical-ai-search-recovery-agency-proposal#webpage","url":"https://groas.com/post/we-took-apart-a-typical-ai-search-recovery-agency-proposal","name":"The AI Search Recovery Proposal I Wouldn’t Sign","description":"A line-by-line teardown of a typical AI search recovery proposal: what to cut, what to keep, and how to judge the work by enquiries instead of traffic and mentions.","inLanguage":"en-US","isPartOf":{"@id":"https://groas.com/#website"},"about":{"@id":"https://graph.groas.com/entity/groas"},"primaryImageOfPage":{"@type":"ImageObject","url":"https://pub-87da24ecbbfc4c3bad6875f3aa013712.r2.dev/generated-images/b08c60ee-1012-4265-9ace-d5bb3f0f8d5c.png"},"breadcrumb":{"@id":"https://groas.com/post/we-took-apart-a-typical-ai-search-recovery-agency-proposal#breadcrumb"},"dateModified":"2026-11-02T06:04:05.677Z","mainEntity":{"@id":"https://groas.com/post/we-took-apart-a-typical-ai-search-recovery-agency-proposal#article"}},{"@type":"BreadcrumbList","@id":"https://groas.com/post/we-took-apart-a-typical-ai-search-recovery-agency-proposal#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://groas.com/"},{"@type":"ListItem","position":2,"name":"Blog","item":"https://groas.com/blog"},{"@type":"ListItem","position":3,"name":"The AI Search Recovery Proposal I Wouldn’t Sign","item":"https://groas.com/post/we-took-apart-a-typical-ai-search-recovery-agency-proposal"}]},{"@type":"BlogPosting","@id":"https://groas.com/post/we-took-apart-a-typical-ai-search-recovery-agency-proposal#article","url":"https://groas.com/post/we-took-apart-a-typical-ai-search-recovery-agency-proposal","mainEntityOfPage":{"@id":"https://groas.com/post/we-took-apart-a-typical-ai-search-recovery-agency-proposal#webpage"},"isPartOf":{"@id":"https://groas.com/blog#blog"},"headline":"The AI Search Recovery Proposal I Wouldn’t Sign","description":"A line-by-line teardown of a typical AI search recovery proposal: what to cut, what to keep, and how to judge the work by enquiries instead of traffic and mentions.","datePublished":"2026-11-02T06:04:05.623Z","dateModified":"2026-11-02T06:04:05.677Z","image":{"@type":"ImageObject","url":"https://pub-87da24ecbbfc4c3bad6875f3aa013712.r2.dev/generated-images/b08c60ee-1012-4265-9ace-d5bb3f0f8d5c.png"},"author":{"@id":"https://groas.com/author/alexander-perelman#person"},"publisher":{"@id":"https://graph.groas.com/entity/groas"},"keywords":"page, search, proposal, pages, work, search recovery, recovery proposal, recovery","wordCount":2205,"inLanguage":"en-US"},{"@type":"Person","@id":"https://groas.com/author/alexander-perelman#person","name":"Alexander Perelman","jobTitle":"Head Of Product @ groas","description":"My background includes Goldman Sachs and Stanford Computer Science.","image":"https://groas.com/media/blog/4f17cac3a81acc224cc1b2cfab8f94bbc295a3089d13e8421125771ef8d4064f.jpg","worksFor":{"@id":"https://graph.groas.com/entity/groas"},"sameAs":["https://www.linkedin.com/in/alexander-433793253/"]},{"@type":"FAQPage","@id":"https://groas.com/post/we-took-apart-a-typical-ai-search-recovery-agency-proposal#faq","mainEntity":[{"@type":"Question","name":"Why shouldn’t an agency promise to restore the site’s old organic traffic?","acceptedAnswer":{"@type":"Answer","text":"Because AI Overviews can answer a search before the reader visits a website, restoring old rankings or session counts may not restore the lost clicks. The proposal should identify which pages still give potential customers a reason to visit and measure the enquiries those pages produce."}},{"@type":"Question","name":"Which lost pages should an agency prioritize for a refresh?","acceptedAnswer":{"@type":"Answer","text":"Prioritize pages that help potential customers decide whether to contact or hire the business, and that can be tied to an enquiry. A page that mainly drew informational visits is not automatically worth rewriting just because it lost traffic; a defensible list of ten buyer pages can be more useful than thirty traffic-led rewrites."}},{"@type":"Question","name":"What GEO work is worth keeping in an AI search proposal?","acceptedAnswer":{"@type":"Answer","text":"Keep work that checks crawler access to relevant commercial pages, keeps offers and dated proof current, makes answers specific, and identifies third-party lists and review pages assistants use. The proposal argues against paying for special AI schema, an `llms.txt` file, or a guaranteed mention count as the core deliverable."}},{"@type":"Question","name":"Who should be responsible for SEO and AI search work?","acceptedAnswer":{"@type":"Answer","text":"One owner should be accountable for the commercial pages, supporting evidence elsewhere, and the path from discovery to enquiry across Google and AI assistants. Separate teams and reports are not enough unless they share responsibility for the same enquiry number and someone checks that the buyer-facing information stays consistent."}},{"@type":"Question","name":"What should an agency’s monthly search report show first?","acceptedAnswer":{"@type":"Answer","text":"It should show enquiries by landing page first, distinguishing commercial and informational pages. Rankings, traffic, branded searches, and citations can help explain changes, but they are supporting evidence rather than business outcomes. Search Console groups AI Overview clicks under the normal Web search type, so it does not neatly separate them."}},{"@type":"Question","name":"When should a business reject a separate GEO retainer or long-term commitment?","acceptedAnswer":{"@type":"Answer","text":"Reject it when the enquiry target is vague or the deliverable is mainly a mention report. Keep a clear, cancellable fee attached to named work and reporting on calls, forms, and booked jobs by landing page, so the business can tell what it is buying before the first invoice."}}]}]}
```