Working on Many Websites Together Without Losing the Thread

The moment a second colleague starts looking at the same domains, search optimisation stops being a craft and becomes a logistics problem. This piece is about the set-up that keeps it bearable: revocable permissions, tags that work as filters, a timeline per engagement, an assistant that consults your own figures, and reports nobody assembles by hand.

Search optimisation software tends to be discussed as though one person sat behind it: somebody opens a chart, draws a conclusion, makes a change. Working alone, that picture holds up.

Life in an Amsterdam agency differs: colleagues covering each other's clients, clients who log in themselves, freelancers riding along for three months. So what the Semalt panel gathers in one place is less a set of charts than a shared shop floor, with campaigns, Google data, market research and indexing side by side.

It is not the tenth website that breaks the system, it is the second colleague

A stubborn belief holds that portfolio trouble comes from the number of domains. Only half true: twenty sites stay in decent shape single-handed, provided your memory is good and nobody has to ask you anything. The strain starts when a second person joins.

From then on, every decision you do not write down is lost. Why a keyword was dropped in February, what was agreed with the client about the English copy — it sat in one head and stayed there.

Nine storage places is no exaggeration: the mailbox, the shared drive, the spreadsheet of links, the task board, the chat group, the meeting notes, the downloads folder, a second spreadsheet holding passwords, the folder of screenshots. None is wrong in itself; they simply cannot be questioned in one movement — exactly what collaboration demands.

The right-hand column is dull, deliberately so: one destination, one searchable whole.

Why you start with access rights rather than with the charts

Most teams build their workspace in the wrong order. The sites go in first, weeks of enthusiastic measuring follow, and only when a client asks to look in does the puzzling over credentials start. One step too late.

That third item is the dullest and the most important. Every workspace older than a year collects spectators: last season's intern, the content marketer who left, the interim manager who stayed three months. Give it a fixed calendar slot — quarterly invoicing is as good a peg as any.

Linking runs through one Google consent screen; Gmail, Search Console and Analytics all come along in that step. That suits clients whose IT department wants to point at every authorisation: rights are granted at a single, documentable moment.

Tags are not folders, they are a switch laid across everything

The word tag suggests a sticker on a file. It behaves differently here: a site tag acts on every view at once, so picking one narrows the whole panel to that selection.

That drives the design. Folders make you think in ownership: which site belongs where. Switches make you think in questions: which cross-section do I want. Four independent axes usually suffice.

The second row is the interesting one in Amsterdam. A consultancy on the Zuidas, a software house in Amsterdam-Noord or a hotel group inside the ring road is searched for in Dutch by local customers and in English by international ones, often on one domain. A single average position then hides that one language version moves while the other stands still. And since Dutch-language results play out differently in Flanders, country must be switchable on its own too.

The temptation of the tree

Almost everyone tries it once: tags that nest inside one another — first the relationship, then the site, then the language version. What you build is a folder structure in a different coat, and it collapses as the portfolio grows.

Do the arithmetic for seventeen domains: four cross-cutting axes come to some fifteen values, a tree to thirty or forty — and the tree still cannot call up every English version on a weekly rhythm at once.

What the team delivers: exports, reports and visuals

Reporting comes before the daily tooling on purpose: it determines how you arrange the rest. Deliver the same document each month and your method shapes itself around it; invent something new each month and you stay in manual labour.

A dump to CSV or JSON runs to ten thousand rows and exists to be calculated with. A PDF is composed server-side and stops at two hundred and fifty rows, because that document exists to be read. The limit irritates at first and is really a favour: it forces you to choose which rows carry this month's story — and that choosing is what a client pays for.

Four shapes are available for the visuals, all deliberately ordinary — a client should not have to decode a chart.

• Movement over time. Answers the question raised at every review meeting.
• Cards with headline figures. One line per quantity, which makes detail screens unnecessary.
• Miniature charts inside the table. A few millimetres of curve behind each row, enough to show movement.
• Heat maps for country and device. They show what an average conceals, such as weak mobile traffic in one country.

Pour the export options built into the workspace into a fixed layout per type of client and only the accompanying text remains. Because background processes keep pulling data in, the figures wait ready when you sit down.

The timeline per engagement: a logbook that keeps itself

My SEO Stream is where all of this converges: one continuous stream per engagement. What is unusual is not its shape but where the entries come from — they are not all written by people.

Such a stream gets long by itself; three provisions handle that. Four buttons sit above the timeline — All, Links, Files, Tasks — narrowing it to one kind of content. Links gives every reference gained this quarter with its values; Files lays out the reports.

Second, every task has exactly three states: active, deferred or declined — an austerity that is deliberate.

Third comes search across the full content of every entry. That sounds obvious and is not: a half-forgotten discussion comes back from one word, with the surrounding messages attached.

The assistant picks its sources before it starts writing

An assistant sits inside the channel, and it differs fundamentally from the language model in your next tab: that model knows the discipline but not your engagement, so its answers are true everywhere and applicable nowhere.

Here something precedes the answer. A separate routing model reads the question and decides which data gets pulled in. Four kinds exist — Search Console figures, standings from the SERP module, the course of the campaign, and blocks you added yourself. Never more than three travel at once, and sometimes the count is zero.

That zero is not a fault. Ask how impressions relate to clicks and your dataset adds nothing. Ask which terms slipped out of the top ten these four weeks and the relevant blocks are consulted. Stacks of keywords or addresses can be handed over in bulk.

The answer appears token by token as it is produced, so you can step in before the model heads somewhere you did not intend. Twenty messages travel with the conversation, so following up needs no re-explaining.

What this looks like in an ordinary working week

By way of illustration, an invented but realistic agency: five people in Amsterdam-West, seventeen domains. Two fintech firms with a Dutch and an English site, a canal-cruise booking platform, three hospitality groups, a design studio, a publisher, two interior webshops, a b2b software company and five smaller service providers.

Wednesday, the half hour everyone joins

The portfolio round sits not on Monday but on Wednesday morning, once the first rush is over. The weekly-rhythm tag brings six domains into view; for those the team checks the headline cards and movement over four weeks. Anything notable enters the relevant timeline as a task with a name on it. The half hour is hard: whatever does not fit is not weekly work.

The other eleven get a monthly pass; in between, the notices appearing in the timelines by themselves are enough.

The hours in between

The real work happens outside that round. A fintech firm reports that its English sign-up form is getting less traffic. The question goes straight into the timeline; the routing pulls in the sources that matter, and two follow-ups produce a usable picture. The first answer, which pointed the wrong way, stays too — in four months somebody will want to see which leads were chased.

At the publisher a proposal goes to deferred, since that section is being rebuilt in the autumn anyway. At one webshop another goes to declined: the platform vendor is already fixing it. Both cost one click and save a recurring argument.

• Someone outside needs to look in. Share that one domain and schedule the withdrawal in the same breath.
• A question that feels familiar. Search the full text first; the answer is usually there already, with its reasoning.
• Something stands out, but there is no time. Record it as a deferred task, not in a chat message that sinks.
• A colleague takes over an engagement. Let them read the timeline back; a handover document goes stale anyway.

The first Friday of the month

Reporting day falls not on the last working day but on the first Friday after it, so the previous month's figures are in; the two-day lag in Search Console data then makes no difference. The layout per client type is fixed, so assembly takes minutes. The bilingual fintech firms each get two documents, one per language version — the language tag does that work.

What does take time is the five sentences atop each report: what these figures represent, and what is about to happen. The one section certain to be read — and human work again, every month.

What a panel never takes off your hands

All of the above concerns the environment the work happens in: where something lands, how it is found again, who may see it. That layer you set up once. The layer above it, never.

That is less exciting than a system promising to take over the thinking, but it is checkable. The gain is a shift of attention: the less time leaks into looking things up and handing things over, the more is left for those four points. Which further parts the environment holds is set out in the full overview of modules; how we use them day to day is described under our services.

Questions teams ask about this

To set this up, work in this order: add the domains and make the Google connection, introduce one tag axis, run one engagement through the timeline for a month, fix one report layout, handle the shares last. Reverse it and you build for a way of working that does not exist yet. More about how we work is on the page about our agency.

Two or three domains are enough to tell whether this suits your team. You can set up a workspace in the Semalt panel and add the rest of your portfolio later.

One final test, and it is not about rankings. Ask a colleague what happened at a randomly chosen client last quarter. If they can show it inside two minutes rather than recount it from memory, your workspace is sound. If not, you need no new measuring instrument but a better place to keep things.