Ninety articles a month across five clients is eighteen pieces per client, or roughly one every working day and a half. Done by hand that is a full-time writer, an editor, and somebody chasing WordPress logins. Done badly with automation it is ninety liabilities published under five other companies' names. SEO automation for agencies only pays when the checking scales with the writing, and that is the part most tooling leaves to you. This is the operating model we built Rankli's Agency plan around: what to set up, what runs without you, and what still needs a person.
What SEO automation for agencies has to solve
Five clients is not one client times five. Each one has its own voice, its own product vocabulary, its own list of words legal will not allow, its own CMS, its own reporting cadence and its own opinion about how much review they want.
Automation that ignores that produces the thing every agency fears: five blogs that read like the same blog. So the unit of configuration has to be the client site, not the account. In Rankli each of the five sites on the Agency plan carries its own site profile, its own keyword set and calendar, its own banned-phrase list, its own publishing connection and its own report. The pipeline is shared. The settings are not.
The second thing it has to solve is review capacity. If every draft needs a senior person to read it before it publishes, your ceiling is that person's reading speed, and no amount of generation changes it. The answer is not skipping review. It is making review cheap by making the machine do the first pass and show its work.
The maths, before you commit to it
The Agency plan is $1,200 a month for 90 articles across five sites, which is $13.33 an article. What you bill for a published, checked, internally linked, live article is your rate card, not ours.
Three costs are easy to forget. Onboarding time per client, which is real but one-off, and which we cover below. Review time per article, which is the number automation actually changes. And the cost of a bad publish, which is not a line item until it happens and is then the only one that matters.
If you would rather start smaller, Growth is $99 a month for 30 articles on one site, and the add-ons take a single site to 60 articles for $85 or 90 for $160. A common way in is one client on Growth for a month, then the Agency plan once the loop is proven. Full detail is on the pricing page.
Week zero: onboarding five clients
Budget two hours per client. It is the highest-return work in the whole system, because everything downstream inherits it.
Voice, from their own writing
Point Rankli at the client's site and it derives a summary, an audience, a voice and tone examples from the pages themselves. Read what it derived and correct it. The most common fix is tone: agencies know a client sounds drier or warmer than their marketing pages suggest, and one sentence of correction here is worth more than editing forty drafts later.
Keywords, then a calendar
Research the phrases, then let the calendar sequence them. Rankli mixes intents deliberately so a month is not eighteen comparison posts in a row, and derives a content type for each title so the client can see the shape of the month at a glance. Eighteen pieces per client per month is enough to cover a cluster properly rather than scatter.
The publishing connection
Ten destinations are supported: WordPress, WordPress.com, Webflow, Shopify, Wix, Ghost, Notion, Framer, a Next.js blog and a signed webhook. Get the credential from the client once, at onboarding, while you have their attention. If their stack is unusual, take the webhook and have their developer accept the payload, or drive the whole thing from the REST API and the CLI.
Banned phrases and the review setting
Ask each client for the words they refuse to publish. Competitor names, regulated claims, the phrase their founder hates. In Rankli that list is a hard check, not a style hint: a draft containing one fails and does not publish.
Then set the review posture per client. A client that trusts the pipeline gets automatic publishing with a check gate. A client in a regulated space gets "hold drafts that fail checks" switched on, and a nervous client gets everything held for approval. Three different postures, same pipeline.
The weekly loop
Once week zero is done, the running cost is about an hour a week per client, and most of it is reading.
Monday, the calendar. Look at what is planned for each client this week. Retitle or reorder anything that has gone stale. This takes ten minutes per client and it is the only planning most weeks need.
Midweek, the checks. Every draft runs through fifteen rules before it can publish: the 1,500 to 3,000 word range, one H1 with no skipped levels, keyword in the title, the opening and an H2, keyword density under 2.5%, the client's banned phrases, an empty-section rule, a rule that the article actually mentions what the client sells, and a fabricated-statistic detector that flags any sentence shaped like a sourced claim without a source. A failed hard rule regenerates the draft once. If it fails again it is held with the reason written on it.
Your job midweek is to read the exceptions, not the ninety. That is the whole change in the economics.
Friday, the report. Traffic comes from each client's own Google Search Console, so the numbers in your deck are the numbers in their account. A day Google did not report is drawn as no data rather than filled with a zero, which is the kind of detail that stops a client meeting going sideways.
What the checks catch before a client does
The value of an automated gate is boring and specific. In practice the rules that fire most are these.
The unsourced statistic. A model writing 2,000 words about a category will reach for a confident-sounding figure with an attribution behind it. Published under a client's byline, that is the single most expensive thing this category can do to you. It is a hard rule.
The empty section. A heading that was planned and never written is the defect a client spots in four seconds, and it is invisible to any scorer that only counts words.
The article about nothing. A piece can hit every structural target and never mention what the client sells. Rankli checks for the client's own vocabulary outside the keyword phrase, and fails the draft when it is missing.
You can run these fifteen checks on any draft, from any tool, in the free SEO article checker. It is a useful thing to put in front of a prospective client who is already using something else.
The first month, day by day
A first month with a new client tends to run the same way, and knowing the shape of it makes the pitch easier.
Days one and two are onboarding: voice, keywords, connection, banned phrases, review posture. Nothing publishes.
Days three to seven are the first four articles. Read all four properly, even the ones that passed. This is where you find the two corrections that improve the other eighty-six: usually a tone note and a product-vocabulary note.
Week two, the drafts start landing without you. Read the exceptions and skim the rest. Expect one or two held drafts; that is the gate working, not the system failing.
Week three, the first internal links start compounding, because there is now an archive to link into. This is also when the client asks whether it is working, and the correct answer is that three weeks is too early to read a search result and you will show them the measurement at the end of the month.
Week four, the report. Search Console traffic, the citation readout, and the list of what was published. If something is not working, this is when you change the keyword plan, not the tool.
By month two the loop is about an hour a week per client, and the interesting question stops being whether the articles get written and becomes which clusters to buy next.
Reporting a client will accept
Two numbers, and they answer different questions.
Search traffic, from their Search Console property, is the one they already understand. Nothing in Rankli estimates it or fills gaps.
Citation share across ChatGPT, Perplexity, Gemini and Claude is the one they are starting to ask about, because their buyers now ask an assistant before they run a search. Rankli tracks a set of the client's buying questions repeatedly across all four engines, and from Growth up it runs as a pre-registered experiment against a held-out control set of questions. That matters because assistant answers drift on their own: in our calibration harness, across 150 independently built worlds in which nothing was done, the naive before-and-after estimator still moved at least 1.3 points half the time. A controlled number is the difference between reporting a result and reporting weather.
Where automation stops
Three things stay with people, and pretending otherwise is how agencies get burned.
Strategy. Which clusters to own this quarter, which client should be publishing less and linking more, when to stop writing and fix a category page. No pipeline decides that.
Client relationships. The month a client's traffic dips for reasons outside anyone's control is a phone call, not a dashboard.
Anything with legal consequences. Regulated claims, comparative advertising, anything a client's counsel would want to see. Hold those drafts, read them, and bill for the reading.
Everything between those, the researching, drafting, checking, linking, publishing and measuring of eighteen articles per client per month, is the part that scales. Our agency page has the per-client setup in more detail.
Start with the $1 trial on one client site: three finished articles in their voice, published live, each with its scorecard attached. Fourteen days to judge the writing, cancel from the app any day, unused articles refunded.