How to run a $5K/month SEO agency with Fable 5.2 + Search Console

Your next SEO client may already rank on page one. Here’s how to spot the clicks they’re missing.

24 min read

Most people who sell Google rankings with AI end up with clients who leave after month two. They sell audits nobody reads. They write reports from memory. They promise rankings they cannot see coming.

This is the 13-step course that turns one laptop into an agency that keeps five clients at $1,000 a month. A model has no idea what Google sees. So the pipeline is the eyes: Fable 5.2 plus each client's own Search Console.

Here is an example of the results you could achieve with this method:

Web designer Geoffrey Merlet reviews adding SEO services for his existing website clients.

Search Console chart shows 103,000 clicks and 12.9 million impressions, with ChatSEO adoption marked.

To get similar results for your business: try ChatSEO free for 7 days

Save this one. It's long on purpose, and you will come back to step 9 every month.

The first screenshot is a real five-star review from August 31, 2026, written by Geoffrey Merlet, a web designer. In short, he says the tool let him add a new source of revenue: he now sells complete SEO services to the clients he already builds websites for, and the audits and optimized texts are generated in a click. That is the whole idea of this guide, one person adding a recurring SEO offer on top of what they already do. The second screenshot is the Search Console curve I put under every link to the tool.

Full disclosure before anything else: I am a ChatSEO affiliate, and every ChatSEO link in this guide is my affiliate link. If you sign up through it, I get a commission. Everything else in here works without it, and I say exactly where it saves you time and where you don't need it.

Who is writing this

I ship apps with AI every day and I document what breaks. Agents, command line tools, automation, the incidents nobody posts about. SEO came into my life the boring way: every app I shipped needed traffic, and every client who saw my apps asked "can you also get us on Google?"

So I built a small, repeatable way to answer yes: a one-person setup (not an "AI SEO empire") that takes a local business or a small site, reads its real Search Console data, finds the pages that are almost winning, fixes them, and reports every month in plain words.

A few honest rules for what follows:

  • $5,000 a month is a goal, not a promise. It is 5 clients at $1,000. I show you where that price comes from, and you will find out in step 6 that most of the market charges less, not more.

  • The Search Console screenshots come from a real small site, a whey protein review site, used here as the demo client. Its owner lets me show the numbers. It is not a client of mine, and I don't claim any of its traffic.

  • Fable 5.2 does the reading, the sorting and the writing. It does not see search volumes, the live top 10 or anyone's backlinks by itself. I say where that matters (step 8) and what to plug in.

  • Every capture in this guide was taken on November 14, 2026.

Why Fable 5.2 changes the math

An SEO agency used to need three people per client: someone to pull the data, someone to decide, someone to write. The data part was a spreadsheet afternoon. The writing part was a week.

Fable 5.2 collapses the first and the last. You give it the export, it reads every row. You give it a page and the query that page is losing on, it writes the new title, the new intro and the internal links. It can run several of those jobs side by side with subagents, each one in its own context, so the client's 600 pages don't drown the one question you asked.

What it does not collapse is the middle: deciding what matters for this business, this month. That stays your job, and it is exactly the part the client pays for.

The other change is cost. At Anthropic's launch prices (step 12 shows the page), a full monthly pass on a small client costs a few dollars of model time, not a few hundred. That is what makes $1,000 a month a fair price for the client and a real margin for you.

Here is the whole agency on one line, so you know where we're going:

Search Console in, quick wins out, pages fixed, a one-page report every month, 5 clients, one person.

Fable can read 600 pages before lunch. It still needs you to tell it which page pays the rent.

Let's build it.

01. The mistake: the free audit nobody reads

Orange robots pull a plug toward a folder beneath the words ‘The setup’.

Every new SEO freelancer starts the same way. They find a prospect, run a free audit tool on the homepage, export a PDF with 140 "issues", and send it.

The PDF lands in Downloads, where it joins 3 printer drivers and a file called final_FINAL_v2.pdf.

Look at what Google shows when anyone, including your prospect, searches "free seo audit":

Google results for ‘free SEO audit’ highlight an AI Overview and free audit tools.

Google's own AI Overview lists the free audit tools at the top of the page, and the tools themselves fill the results under it. Your prospect can get the same 140-issue PDF in ten seconds, without you. So a generic audit is not a gift, it's noise. Worse, it tells the client "SEO is a list of technical errors", and the next freelancer with a longer list wins.

A crawler can find 140 errors. Your client pays you to pick the first fix.

Here’s what that decision looks like: these 3 pages, these 5 queries, this month, because the data says so.

Three things a generic audit cannot do, and you can:

  1. Read the client's own Search Console, the only place where Google says which searches already show the site and how often.

  2. Rank fixes by money, not by severity. A missing alt text on a blog image is "an error". A service page sitting at position 8 for the client's main service in their city is a lost phone call every day.

  3. Come back next month with the same numbers and say what moved.

Everything in the next 12 steps is built around those three. Keep the audit tool for the 5% of sites that are truly broken (no index, no HTTPS, a blocked robots file), and never lead with it.

Paste this in Fable 5.2 the first time you look at a prospect, before you write a single line to them:

You are my SEO lead. I am looking at a local business website as a prospect: <URL>.
Do NOT give me a generic audit. Answer only these questions, briefly:
1. What does this business sell, to whom, and in which city or area?
2. Which 5 searches would a ready-to-buy customer type to find it? Write them as a customer would.
3. Which page of the site should rank for each of those 5 searches? If no page fits, say "missing".
4. What is the single most obvious problem on the page that matters most (title, intro, missing service page, no city named)?
Keep it under 200 words. Say clearly what you could not verify without Search Console data.

That last line matters. Fable will tell you what it is guessing. Write it down: those guesses are what the Search Console access in step 3 will confirm or kill.

02. The fast track: skip the setup if you want to start today

Steps 3 to 5 build the whole agency in Claude Code, with your own folders and subagents. It takes an evening, and I recommend you do it once, because you will understand every moving part.

But if you want to show a prospect something useful tomorrow morning, there is a shortcut: a tool that already plugs into Search Console and live SEO data, and runs these workflows as one-click playbooks.

That's the one I use, ChatSEO. Reminder from the top: I am an affiliate and the link below is my affiliate link.

ChatSEO homepage highlights the website address field used to start an analysis.

You paste the client's site, connect their Search Console, and ask for the quick wins. It does steps 7, 9 and 11 of this guide with live keyword volumes, the live SERP and backlink data that Fable cannot see by itself. The first run is free, so you can test it on a real prospect before you decide anything.

When to use the fast track:

  • You have a prospect call this week and nothing set up yet.

  • The client cares about keyword volumes and competitors, and you don't want to wire those data sources yourself.

  • You want the report done in minutes, then you spend your time on the decision and the call.

When not to bother: if you love building your own stack and you have an evening, the next three steps give you a setup you own completely. Many people do both: the fast track for prospects, the own stack for the monthly work.

Search Console chart marks ChatSEO adoption before the rise in clicks and impressions.

To get similar results for your business: try ChatSEO free for 7 days

03. The install: the client's Search Console, on day one

Everything in this agency runs on one access: the client adds you to their Google Search Console property. Until that invite arrives, asking Fable where the clicks went is like asking your GPS for directions with location turned off. Accepting the invite takes two minutes, and it changes the whole relationship, because from now on you both look at Google's own numbers, not your opinion.

Here is Google's help page for it:

Search Console Help highlights full user access rather than ownership for an SEO consultant.

What to ask for, word for word:

"Could you add me as a Full user on your Search Console property? Settings, then Users and permissions, then Add user, with my email. It's read access to your Google data, it doesn't let me change your site."

Why Full user and not Owner: an owner can add and remove other people and change settings. You don't need that, and asking for it makes a careful client nervous. Google's page says a full user has view rights to all data and can take some actions, which covers everything in this guide.

If the client has no Search Console at all, that is your first paid task, and it is a good one: you verify the domain property (a DNS record, five minutes with their host), submit the sitemap, and in 2 to 3 days you have the first data. Tell them the truth: the history starts the day you connect it, so month one is a baseline.

Once you have access, export the data for Fable. In Search Console, open Performance, pick the last 3 months, then Export, and take both the Queries and the Pages tables. Two CSV files. You will drop them in the client's folder in step 5.

If you want to automate the export later, the Search Console API gives the same tables. Don't start there. Start with the button. An agency that works by hand for its first two clients knows exactly what to automate for the third.

The prompt for the first look at a fresh export:

I am attaching two Google Search Console exports for my client <business>, last 3 months:
queries.csv (query, clicks, impressions, CTR, position) and pages.csv (page, clicks, impressions, CTR, position).
1. Summarize the site in 5 lines: total clicks, total impressions, average position, and what kind of searches it shows up for.
2. List the 10 queries with the MOST impressions and the FEWEST clicks. For each one: impressions, clicks, position, and the page that ranks.
3. Flag any query where the position is between 4 and 15: those are the closest to page-one clicks.
Use only the numbers in the files. Do not estimate search volumes.

04. The crew: 7 subagents and one director

One long chat with Fable gets messy fast: by the time you ask for the report, its context is full of CSV rows and drafts. The fix is the one Claude Code ships for exactly this: subagents.

Claude Code documentation highlights how to create custom subagents with separate contexts.

Anthropic's docs say it plainly: a subagent runs in its own context window, with its own system prompt and its own tool access, does the side task, and returns only the summary. That's what an agency crew is. Each member does one job and hands back a short result to the person who decides.

Here's the crew I run for every client:

  1. The director: reads the brief, decides which jobs run this month, and checks every result before it reaches the report. This is the main session, the one you talk to.

  2. The researcher: reads the client's site and the brief, and writes down what the business sells, to whom, where.

  3. The GSC analyst: reads the two exports and returns the quick-win list (step 9).

  4. The writer: drafts new pages when a search has no page (step 8).

  5. The title fixer: rewrites titles and intros of pages that rank but don't get clicked (step 10).

  6. The linker: proposes internal links from strong pages to the pages you are pushing (step 10).

  7. The reporter: turns this month's numbers into the one-page report (step 11).

Seven is not magic. It's the number of distinct jobs I found in my own month. Start with three (analyst, title fixer, reporter) and add the rest when you feel the need.

One warning from the same docs page: each subagent sends its own requests, and they count toward the same usage limits as your main conversation. Running seven of them on every small question burns your budget for nothing. Call the crew when the job is real, not to say hello.

To create one, Claude Code reads a markdown file per subagent from the project's .claude/agents/ folder. The GSC analyst, for example:

---
name: gsc-analyst
description: Reads the client's Search Console exports in gsc/ and returns the quick-win list. Use it whenever the director needs to know which queries or pages to work on.
tools: Read, Grep, Glob
---
You analyse Google Search Console exports for a small business.
Read gsc/queries-*.csv and gsc/pages-*.csv for the latest month.
Return at most 10 quick wins: queries with many impressions, few clicks, and a position between 3 and 20.
For each: the query, impressions, clicks, CTR, position, the page that ranks, and ONE fix in one sentence.
Use only the numbers in the files. Never invent a search volume. If the data is too thin (under 500 impressions in total), say so and stop.

Notice the tools line: the analyst can read files and nothing else. It can't edit the site, it can't browse. A crew member with fewer powers makes fewer surprises.

05. The folder: one folder per client, nothing shared

This is the boring step that makes the agency possible. Every client gets one folder, and everything about that client lives in it: their brief, their crew, their data, their pages, their reports.

Annotated terminal tree separates agent instructions, Search Console exports, and the client brief.

That's a real tree of the folder I use as a template, with a fictional roofing company as the client:

  • CLAUDE.md is the file Claude Code reads first in that folder. It holds the client's rules: their tone, what they never want on their site, who approves what.

  • brief.md is the business in one page: services, cities, prices if public, the 5 searches that make them money, their top 3 competitors.

  • .claude/agents/ holds the 7 subagent files from step 4. Copy them from client to client; only the brief changes.

  • gsc/ holds the monthly exports, named by month, never overwritten. Month 3 compares with month 1 only if month 1 is still there.

  • pages/ holds the drafts you write for them, one file per page.

  • reports/ holds one report per month.

  • quick-wins.md is the running list from step 9, with a date and a status next to each line.

Why one folder per client and not one big project: context. When you open Claude Code in the roofer's folder, Fable sees the roofer and nothing else. No plumber keywords leaking into a roofing title, no wrong city in an intro. It also makes offboarding trivial: a client leaves, you zip their folder and hand it over. Clients love that, and it's the honest thing to do.

The CLAUDE.md I start every client with:

Website: <URL>  |  Area: <city / region>  |  Language: <language>
Contact who approves changes: <name>, by email. Nothing goes live without their written OK.

## What they sell
<3 to 5 lines from brief.md>

## Rules
- Use only data from gsc/ and from the live site. Never invent figures, reviews or prices.
- Every page or title change is a proposal in pages/ or quick-wins.md first, with the reason and the number behind it.
- Write like the client talks to customers: <plain / friendly / formal>. No hype words.
- Never promise a ranking or a date.

## This month
<the 1 to 3 goals you agreed, updated on the 1st>

06. The offer: 5 clients at $1,000

An orange robot holds a dashboard above miniature houses under the heading ‘The money’.

Now the money. $5,000 a month is 5 clients at $1,000, or 10 at $500. I prefer 5 at $1,000: fewer calls, more time per site, and the client takes you seriously.

Is $1,000 a fair price? Here is the best public answer I know, an Ahrefs survey of 439 SEO professionals:

Ahrefs pricing survey highlights the most common monthly SEO retainer of $500 to $1,000.

Their key takeaways: SEO costs $2,917 a month on average, 23% of SEOs charge $500 to $1,000 a month, and 78.2% charge a monthly retainer. The survey is from 2024 (updated August 15, 2024), so take the exact figures with a grain of salt, but the shape holds: $1,000 a month sits in the most common band, well under the average. You are not the cheap option and not the luxury option. You're the normal price, with more data on the table than most.

What the client gets for $1,000 a month, written exactly like this in your proposal:

  1. Their Search Console, read every month, by someone who knows what the numbers mean.

  2. Up to 5 fixes a month on the pages closest to bringing calls: titles, intros, internal links.

  3. One new page a month for a search they're missing (a service, a city, a question their customers ask).

  4. One page of report on the 1st of the month: what changed, what it did, what's next.

  5. A 20-minute call a month, optional.

What you never put in the offer:

  • A guaranteed ranking. Nobody can promise one, and the client who believes you will leave angry in month 3.

  • "Unlimited" anything. Five fixes a month is a promise you can keep for 5 clients.

  • Backlink packages. Buying links is a risk you take with the client's domain, not yours.

On the contract: month to month, cancel any time with 30 days' notice. It sounds scary. It's the reason they sign. The first three months decide whether they stay, and the report in step 11 is what makes them stay.

07. The first client: a free quick-win list, not a free audit

Remember step 1: the generic audit is noise. Here is what you give for free instead, and it gets answers because it is about their money, not about errors.

The free quick-win list: three pages of their site that Google already shows a lot, with almost nobody clicking, and the one fix for each. You can only make it with their Search Console data, which is exactly why it is worth something, and why it's a natural reason to ask for access.

Here is what the raw material looks like on the demo site, the Pages table sorted by impressions over 3 months:

Search Console flags a query with 2,451 impressions, zero clicks, and an average position near eight.

Read the second row. The "best whey protein for bulking" page was shown 2,451 times in 3 months, at an average position of 8.0, and got 0 clicks. Position 8 is the bottom of page one. Google already likes this page. The searchers just don't pick it. That's quick win number 1 on the free list, and the fix is almost always the title and the first lines (step 10).

Compare with the first row: 2,514 impressions too, but at position 59.7. That page is on page six. No title rewrite will fix that this month. It goes on the "later" list, not the free one.

So the rule for the free list: high impressions, low clicks, and a position you can actually move, roughly 3 to 20.

How to get the data from a prospect who is not yet a client: you ask for the Search Console access in the first message, as the condition for the free list. "Give me read access for a week, I'll send you the 3 pages closest to bringing you more calls, with the fix for each. If you don't like it, remove me." Most small business owners have never had anyone look at that data. A good part of them say yes.

Where do you find those prospects? Step 13 covers it. For now, the prompt that turns an export into the free list:

Here are the Search Console exports (pages.csv and queries.csv, last 3 months) of <business>, a prospect.
Build a FREE quick-win list of exactly 3 pages:
- Only pages with at least 300 impressions, a CTR under 2%, and an average position between 3 and 20.
- For each page: the URL, impressions, clicks, position, the main query that triggers it (from queries.csv), and ONE fix in one sentence.
- Rank them by what a new click is worth to this business (a service page beats a blog post).
Then write a 120-word email to the owner presenting the 3 fixes in plain words, no jargon, no promises of rankings, signed <my name>.

Read the email before you send it. Fable writes well, but you know the person.

08. Search intent: one page per search, and what Fable can't see

The second fix a small business always needs: a page for each thing a customer searches. One "Services" page listing 12 services can't rank for 12 different searches.

Look at what Google does with a search like "emergency plumber chicago":

Chicago emergency plumber results show business listings and a map above an organic result.

Before any website, Google shows a map with three businesses, their ratings, "Open 24 hours" and the phone number. Someone typing that search wants a person on the phone, now. The first classic result under the map is a page whose title says exactly the search: a 24-hour plumber in Chicago.

That's search intent. For this search, a client needs two things: a complete Google Business Profile (that's the map), and a page titled for that exact service and city, with the phone number in the first screen. A 2,000-word blog post about "how to fix a leaking pipe" answers a different search, and it will never rank here.

So for every client, you build a small table: the searches that make them money, the intent behind each (call now, compare prices, learn something), and the one page that should answer it. If that page doesn't exist, it's next month's new page (the 3rd line of your offer).

Here's the honest part: Fable 5.2 cannot see the live Google results by itself, it cannot see how many people search each phrase every month, and it cannot see who links to which page. It will happily guess all three if you let it. Don't. Three ways to fill the gap:

  1. Open the search yourself, like I did above. For local services, 5 searches by hand tell you more than any tool.

  2. Use the client's Search Console: if a query has impressions, people search it. It's the only volume data that is about their site.

  3. Plug a data source in: a keyword tool's export, or a tool that has live SERP and volume data built in (the fast track in step 2 does this).

The prompt for the intent table, with the guard against guessing:

Using brief.md and gsc/queries-<month>.csv, build a search intent table for <business>.
Columns: search | intent (call now / compare / learn) | the page that should answer it | does that page exist? (URL or "missing") | evidence.
Rules:
- Only use searches that either appear in the queries file (give impressions) or that I list here: <searches I checked by hand>.
- In "evidence", write "GSC: N impressions" or "checked by hand". Never write a monthly search volume you did not get from a file.
- Mark the 2 missing pages that matter most for revenue.

09. Quick wins: impressions up, clicks down

This is the step you will run every single month, for every client. It's the engine of the agency.

Open Performance in Search Console, switch to Queries, and sort by impressions, highest first:

Search Console’s impression-sorted query table highlights a high-visibility query with zero clicks.

On the demo site, the top query is "whey protein price comparison": 2,782 impressions in 3 months, an average position of 5.7, and 0 clicks. Position 5.7 is the middle of page one. People see the result almost 3,000 times, and nobody clicks it. That's the definition of a quick win: Google already did the hard part, and the page loses at the very last step, the click.

Why sort by impressions and not by position? Because position alone lies. A query at position 2 with 10 impressions is worth nothing. A query at position 9 with 3,000 impressions is a page that is one good title away from real traffic. Impressions tell you where the demand already is, for this site. Position tells you if you can move it this month.

The quick-win filter I use, in plain words:

  • Many impressions (the top of the sorted list).

  • Few or no clicks, so a CTR well under what the position should get.

  • A position between about 3 and 20: close enough to page one that a better title, a better intro or a few internal links can move it.

  • A business query, one that could bring a sale, a call or a lead. "whey protein price comparison" is a buyer comparing prices. That's money. A random question with no buying intent goes lower on the list, even with more impressions.

The second query on the demo site, "whey pre bed", has 1,686 impressions at position 25.2. Further down, page three. It goes on the list for later, with a note: it needs a better page, not just a better title.

Every month, the GSC analyst from step 4 runs this on the fresh export and updates quick-wins.md. Each line gets a date, the numbers on that date, the fix you chose, and a status (proposed, approved, live). Next month, you check the same lines first. That's how you prove what moved.

The monthly prompt, for the director to hand to the analyst:

Use the gsc-analyst subagent on gsc/queries-<this month>.csv and gsc/pages-<this month>.csv.
Then compare with quick-wins.md:
1. For every line already in quick-wins.md with status "live": the impressions, clicks and position then and now.
2. Up to 5 NEW quick wins (business queries, many impressions, few clicks, position 3 to 20), each with one fix.
3. Anything that dropped by more than 3 positions since last month.
Write the result into quick-wins.md (append, never delete a line) and give me a 5-line summary.

10. Titles and internal links: the cheapest fixes

Most quick wins are fixed by two things that cost nothing: the title Google shows, and the links from the client's other pages.

Titles first. Google's own documentation is clear that the title link in a result comes from several sources, mainly the page's title element, and that it may rewrite a title that is vague, too long or doesn't describe the page:

Google Search Central illustrates title links using ‘How to make crispy fried eggs’.

Their first best practice: every page should have a title in its title element, descriptive and concise. Sounds basic. Half the small sites I open have "Home" or the business name alone on their main service page.

For the demo quick win, a page sitting at position 5.7 for "whey protein price comparison" with zero clicks, the question is: does the title say "price comparison"? Does the snippet show a price? If the searcher wants to compare prices and the title says "Our whey protein guide", they skip it. The fix is a title that repeats the search in human words and promises what they want, plus a first paragraph that delivers it in the first two lines.

The rules I give the title fixer:

  • Put the words of the search in the title, early, the way a person would say them.

  • Promise the thing the intent wants: a price, a comparison, a phone number, an answer.

  • Keep it short enough to show fully, around 50 to 60 characters is a safe habit.

  • Never stuff the city three times or add "Best!!!". Google rewrites spammy titles anyway.

  • Change one page's title at a time and note the date in quick-wins.md, so you can see what moved.

Then internal links. A page that the site itself never links to looks unimportant, to visitors and to Google. The linker subagent reads the client's pages and finds the strongest ones (the homepage, the pages with the most clicks in the Pages export) that talk about the topic, and proposes one sentence with a link to the page you are pushing. Two or three good links from relevant, visited pages often help a page climb the last few positions.

The prompt for both, on one quick win:

Quick win: the page <URL> ranks at position <P> for "<query>" with <N> impressions and <C> clicks (last 3 months).
1. Read the live page. Propose 3 new titles (max 60 characters) that use the words of the query and promise what the searcher wants. Say which one you'd pick and why.
2. Propose a new first paragraph (max 60 words) that answers the query in the first two lines.
3. From pages.csv, pick the 3 pages with the most clicks that are related to this topic, and for each, write one sentence I can add with a link to <URL>, with its anchor text.
Write everything to pages/<slug>-fix.md for the client's approval. Change nothing live.

Notice "change nothing live". The client approves, then you (or their web person) make the change. In month one, that rule is what builds trust.

11. The monthly report: one page, three numbers

The report is why clients stay: not the work, the report about the work.

Most agency reports are 30 slides of charts nobody understands. Yours is one page the owner can read on their phone. It starts with the line they care about:

Annotated Search Console report highlights 58,100 impressions and the period covered by the report.

That's the demo site over 3 months: 518 clicks, 58.1K impressions, a 0.9% CTR and an average position of 32.9. Look at the curve: impressions climb from early October and stay up. That's what you point at, the month you report on, compared with the month before.

The one-page report, always in the same order:

  1. Three numbers, this month against last month: clicks, impressions, and the number of business queries on page one. With the arrow up or down. No average position on the first line, it confuses owners (it goes down when the site shows up for more searches, which is good news that reads like bad news).

  2. What we did: the fixes that went live, each with its date. Three to five lines.

  3. What it did: for each fix from the previous months, its query's numbers before and after. Honest. If a fix did nothing yet, say "no change yet, we check again next month".

  4. What's next: the 3 to 5 quick wins for this month, and the new page you'll write.

That's it. One page. The owner reads it in two minutes and knows exactly why they pay you.

The reporter subagent writes it from the data, you read it, you fix one or two sentences, you send it on the 1st. Clients notice when a report arrives on the same day every month. It's the cheapest trust you can buy.

Use the reporter subagent. Inputs: gsc/ exports for <this month> and <last month>, quick-wins.md, and the list of changes that went live this month.
Write reports/<YYYY-MM>-report.md, one page, for the owner of <business>, who is not an SEO:
1. Clicks, impressions, and number of business queries in the top 10: this month vs last month, with the % change.
2. What we did this month (dated, one line each).
3. Results of past fixes: each one's query, position and clicks before vs now. If nothing moved, say so plainly.
4. Next month: 3 to 5 actions, one line each, with the reason in plain words.
No jargon, no average position in part 1, no promises. Under 300 words.

12. The money map: what one client really costs you

Now let's see if $1,000 a month leaves you anything.

The main cost is model time. Here's Anthropic's launch page for Fable 5.2, with the price:

Highlighted model pricing paragraph identifies token costs used to estimate the monthly cost per client.

Fable 5.2 costs $8 per million input tokens and $40 per million output tokens, on the API. There is an optional fast mode at $16 and $80, which you don't need for a monthly report.

A rough month for one small client, by my own estimate (measure yours, it depends on the size of the site):

  • Reading the two exports and the brief a few times: maybe 1 to 2 million input tokens across the crew.

  • Writing 5 fixes, one new page and the report: maybe 100,000 to 200,000 output tokens.

At list price that's roughly $8 to $16 of input and $4 to $8 of output. Call it $15 to $25 a month per client, even if you are generous. If you use Claude Code on a subscription instead of the API, it comes out of your plan's usage instead, and step 4's warning applies: subagents count toward the same limits.

Your other costs, the ones people forget:

  • Your time. Plan about 6 to 8 hours per client per month at the start (data, decisions, the call, the report), and about 4 once the folder and the crew are in place. Five clients is roughly one week of work per month. That's the real reason to stop at 5 before hiring.

  • A data tool, if you want volumes, live SERPs and competitors without wiring them yourself (the fast track in step 2, or any keyword tool). Count it once, not per client.

  • Your contract, invoicing and the occasional client who pays late. Normal business stuff.

The money map on one line: 5 clients x $1,000 = $5,000 a month in revenue, minus maybe $100 of model time, minus your tools, and about a week of your month.

The goal is not the $5,000. It's the shape: recurring revenue, one week of work, the client's data on the table every month. That's an agency you can keep.

13. The 30-day launch: from zero to the first paying client

Here's how I'd go from nothing to the first client in 30 days, if I started today.

Where the prospects are: every local search shows them to you. Search your city plus a service, like "roofing company austin":

Austin roofing search results highlight three map listings and the ‘More businesses’ button.

The three businesses in the map own the visibility: 5.0 stars, hundreds or thousands of reviews. They are not your first prospects. Your prospects are under "More businesses": the good companies at 4.7 stars with a weak website, no page for their main service, or a homepage titled "Home". They already have customers who love them. They are invisible on Google for reasons you can fix.

Week 1: the setup.

  • Build your template folder (step 5) and your 3 first subagents (step 4).

  • Do the full routine on one site you can access: your own, a friend's, a family business. That's your demo, with real screenshots, like the ones in this guide.

  • Write your one-page offer (step 6).

Week 2: 20 prospects.

  • Pick one service and one city you know. Open 5 searches by hand, list 20 businesses from the "More businesses" pages.

  • For each, run the prospect prompt from step 1 on their site. Keep the 10 with the clearest problem.

  • Send each of the 10 a short, personal message: what you noticed on their site, and the offer of the free quick-win list in exchange for read access to their Search Console.

Week 3: the free lists.

  • Every yes gets the free list within 48 hours (step 7). Three pages, three fixes, plain words.

  • Offer a 20-minute call to walk through it. On the call, you don't pitch. You explain the list. At the end: "I can do this every month, with the report, for $1,000. Month to month."

Week 4: the first month.

  • The first client signs. You do the month: fixes, one new page, and the report on the 1st (step 11).

  • You ask them, after the first report, if they know another business owner who'd want the same. Local business owners all know each other.

Most people won't sign in the first 30 days. That's fine. The routine is the same in month two, you just have more lists out. The goal for day 30 is not 5 clients. It is one client, one report sent on time, and a folder you can copy.

Here's the prompt that prepares week 2 for you:

I'm launching a small SEO service for <service type> businesses in <city>.
Here are 20 businesses I found in Google's "More businesses" list, with their websites: <list>.
For each website, in one line: does it have a dedicated page for <main service>? Does its homepage title name the service and the city? Is a phone number visible in the first screen?
Then rank the 10 with the clearest, most fixable problem, and for each write a 60-word personal message to the owner: one specific thing I noticed, and the offer of a free list of 3 quick wins if they give me read access to their Search Console for a week. Friendly, no jargon, no promises.

The whole agency in one picture

If you skipped to the end, here's the course on one screen:

  1. The mistake: free audits are noise. Sell decisions, not error lists.

  2. The fast track: a ready tool if you need results this week.

  3. The install: Full user access to the client's Search Console, day one.

  4. The crew: 7 subagents, one director, each with one job.

  5. The folder: one folder per client, with its brief and its rules.

  6. The offer: $1,000 a month, 5 fixes, 1 new page, 1 report, month to month.

  7. The first client: a free quick-win list from their own data.

  8. One page per search intent, and never let Fable guess volumes.

  9. Quick wins: sort by impressions, fix the pages with few clicks.

  10. Titles and internal links: the cheapest fixes there are.

  11. The monthly report: one page, three numbers, on the 1st.

  12. The money map: a few dollars of model time per client, a week of your month.

  13. The 30-day launch: 20 prospects, 10 free lists, one client.

Fable 5.2 does the reading and the writing. Search Console gives you the truth. You make the decisions and keep the promises. That's the agency.

What I didn't test

  • I didn't run this exact 13-step routine on 5 paying clients this month. The steps come from my own builds and from the demo site you saw. Your numbers will differ.

  • The token costs in step 12 are my estimate for a small site, not a measured bill. A 5,000-page e-commerce site will cost more. Measure your first month.

  • The Ahrefs pricing survey is from 2024. Prices in your country and niche may be different. Ask a few local agencies what they charge before you set yours.

  • Quick wins move in weeks, sometimes months, not days. Some won't move at all. That's why the report says so honestly.

If you build it, tell me what broke. That's the part I document.

And if you want the shortcut for steps 7, 9 and 11, with live volumes, SERPs and backlinks on top of Search Console, it's the tool I use (affiliate link, as said at the top):

ChatSEO chart marks the adoption date alongside rising search traffic and 11,700 total clicks.

To get similar results for your business: try ChatSEO free for 7 days


Building an SEO agency is mostly about automating the grunt work (data pulling, writing) and keeping the decisions human. The demo-vs-product checklist in the welcome kit shows the exact infrastructure moves that separate a one-month client from a repeatable $1K/month relationship.

→ Get the welcome kit