Matomo, Plausible or Google Analytics: which fits your website?

Custom events record the steps that matter

A page view tells you very little about whether someone can use a service. For a configurator, you might need to record the start, each step and the completed configuration. For an internal search, knowing which queries returned no results is often more useful than counting searches.

A contact form should count as complete only after the enquiry has been successfully processed. Clicking “Send” can still lead to an error message.

For each custom event, we define its trigger, name and properties. The application reports it through an agreed interface, often a data layer. That way, measurement does not depend on a button keeping the same label or position after a redesign.

Names, email addresses and message contents do not belong in these events. URLs and search terms can contain personal information too. We send only what the agreed reporting needs.

We test events through the real user journey, including on mobile and with repeated clicks. Each completion needs to arrive at the right point, exactly once.

Connecting JENTIS, tag management and consent

JENTIS collects data, processes it on the server and forwards it to specified recipients. The analytics platform then handles the reporting.

This can be useful when several marketing services need the same events, or when data needs to be reduced before it is passed on. For a small website with just a few things to measure, the extra work is often hard to justify.

Either way, we need to know which service receives each event. If the browser and server report the same completion, the integration must prevent double counting. The choice a visitor makes in Usercentrics, Cookiebot or another consent platform also has to reach the point where processing happens. Blocking a browser tag does not stop a server from forwarding data.

Server-side tracking does not replace consent where it is required. The permitted processing has to be established for the actual setup. Our article on tracking, privacy and cookie banners covers that technical and legal coordination in more detail.

What the other tools contribute

Search Console, Otterly and JENTIS do different jobs from Matomo or GA4. They complement website analytics rather than replace it.

Search Console and Otterly show what traffic reports miss

If a service page appears frequently in Google but gets few clicks, we look at the queries, its position, title and description. If people reach the page but few suitable enquiries follow, we look at the offer and the route to making contact.

In June 2026, Google introduced dedicated reports for its generative AI search features. These include impressions for URLs shown in AI Overviews and AI Mode. An appearance in an answer is not yet a visit to the website.

Shared dimensions such as page and date let us bring this data together with website analytics. They do not let us connect an individual search query to a later project enquiry.

Otterly regularly puts a chosen set of questions to AI services. It records whether a company is mentioned, which sources are linked and which competitors appear. That shows trends within the monitored questions, not a brand’s total reach across every conversation with an AI assistant.

It is a useful addition to SEO and GEO. An article may be cited in an AI answer without anyone clicking it. Your traffic report will not show that visibility.

When AI agents visit your website

Not every visit associated with AI means the same thing. There are three distinct cases:

• A person follows a link in an AI answer. Where referral information is available, analytics can attribute the visit to that service.
• An agent uses a browser on someone’s behalf. If it runs the tracking script, its actions can appear in your reports.
• A crawler or chatbot requests content directly from the server. Without running the tracking script, it remains invisible to browser-only measurement.

Matomo already distinguishes these kinds of AI traffic. What it captures depends on the configuration and the technical signals available. Direct requests also need data from the server or a service in front of it, such as Cloudflare. Not every automated request can be reliably identified.

Marketing wants to know whether links from AI answers bring useful enquiries. Operations needs to know whether bots can reach important content, whether the firewall blocks them and whether automated requests create unnecessary load. A bot fetching a hundred pages is not a hundred prospects. Nor does a fetch prove that the page will appear in an answer.

Otterly’s Agent Analytics can examine those bot requests using your own access data, connected through server logs, a webhook or Cloudflare. A bot request, a later citation and a website visit still do not add up to a traceable end-to-end customer journey.

Custom dashboards and automated reporting

APIs let us bring data into an existing reporting tool or build a dashboard around a specific need. First, we check which metrics are available, how fresh they are and what request limits apply. Not every report in a product’s interface is available through its API.

An editorial dashboard could put Google clicks, citations in monitored AI answers and completed contact goals alongside each article. Sales may care more about the services people ask for and which leads remain relevant after review in the CRM. Before building anything custom, we check whether an existing report or connector already does the job.

Data from different sources needs a common basis. One system may include campaign parameters in a page URL while another leaves them out. Date ranges and time zones may differ. We agree how to handle those differences and show when each source was last updated. Search impressions, visits and bot requests remain separate measures. Adding them together would not produce a meaningful reach figure.

Consent and campaign attribution can also change how conversions are counted. We compare the measurement with the orders or enquiries actually received by the shop or form. Changes to tracking are documented so that comparisons over time still make sense.

That gives us a basis for useful automation:

• A weekly report summarises important pages, with a comparison period and links to the underlying reports.
• A sustained fall in Google clicks to a key service page notifies the person responsible.
• If enquiries keep arriving through a form but its analytics event stops, a task is created to check the tracking.

We agree minimum volumes and time windows for alerts, so that ordinary fluctuations do not trigger a stream of messages. Missing or late data must not turn into zero visits in a report. If a request fails, the workflow reports the failure rather than inventing a drop in performance.

For reporting, we use read-only access limited to what is needed. Secret API keys stay on the server, out of the website code sent to visitors. We define what a dashboard or workflow may receive just as carefully as permissions in the original tool. If a workflow is also allowed to change campaigns, budgets or published content, that needs its own rules and approvals.

Matomo MCP lets an AI assistant work with your analytics

Matomo released its official MCP server in May 2026. MCP stands for Model Context Protocol. It gives a compatible AI assistant a way to request data from Matomo. Instead of assembling several reports, you can ask a specific question:

“Compare the campaigns from the past four weeks by successfully submitted enquiries. Separate mobile and desktop traffic, and show visits and enquiries alongside the conversion rate.”

That only works if enquiries are recorded as goals and campaigns are correctly attributed. The connection cannot recover events you never collected. It makes existing data easier to use.

MCP is included in Matomo Cloud. On-Premise installations need the MCP Server plugin. In both cases, the feature must be explicitly enabled and access configured. The MCP server passes requests to Matomo’s API and returns the data. The AI assistant interprets it, not Matomo.

Google Analytics and Piwik PRO also offer their own MCP servers. Plausible has a community option. Their permissions differ: Google’s server reads data, while Piwik PRO can also make changes once that capability is separately enabled. For analysis, we start with read-only access.

In Matomo, we use a dedicated account limited to the required websites. There is another risk that is easy to miss: URLs, page titles and campaign names can contain malicious instructions. An assistant must treat them as data, not commands.

Self-hosting Matomo does not keep an external AI assistant inside your infrastructure. Once it retrieves the data, that data is processed outside your installation. Before enabling access, we agree what may be sent, who receives it and the conditions for storage and further use. Aggregated figures are enough for most questions. Individual visit histories usually add nothing useful.

With AI-assisted analysis, we always ask to see the underlying numbers, dates and filters. Two enquiries instead of one is a hundred per cent increase. That alone tells us very little about a campaign.

Who will act on the findings?

Once it is set up, analytics needs a place in the working week. A marketing team running campaigns will check it more often than a business whose website mainly explains its services. Both need someone who turns a finding into a task and comes back to see whether the change helped.

The tracking itself remains part of ongoing website maintenance. A new form, a changed consent setting or an additional service can affect existing reports. Accounts and permissions need attention too, alongside the other parts of web security.

We work with marketing, IT and privacy specialists to agree what the reporting should answer, how the data is collected and who will use it. Let’s talk about the decisions your analytics needs to support and what is missing today.