What a private search engine actually has to give up

Anyone can put the word private on a search box. The more useful question is which features a search engine gave up to earn it, because personalisation and privacy genuinely trade against each other. Here is the trade, itemised.

No search history tied to you

Marisei does not keep a per-person log of what you searched for. There is no query history page, because there is no query history to show you. That also means we cannot offer 'resume where you left off', autocomplete built from your past searches, or recommendations drawn from your interests — real features we do not have and cannot add without breaking the premise.

The practical test of a privacy claim is what a company could hand over if asked. A per-person search log is the single most revealing thing a search engine can hold, so the only durable answer is not to assemble one.

No behavioural profile, so no personalised ranking

Results are ordered by what fits the query and, in local search, by distance from the area you chose. They are not reordered based on what someone like you clicked last week. For most practical searches — a tyre shop, a camera store, an opening time — this makes almost no difference, and the ordering is easier to reason about because it depends only on the query in front of you.

It does mean two people searching the same words in the same place see the same page. That is a feature if you are trying to check what your customers see, and a limitation if you were hoping the engine would learn your preferences.

What we do measure, and why

We measure aggregate usage — which pages get visited, how many searches complete, how often the trust assessment times out, which providers are erroring — because a product with no measurement is a product that quietly breaks and nobody notices. That measurement is gated behind consent and is aggregate, not a per-person timeline.

There is also an operational log of the trust checks themselves: which domains were assessed, how long each took, and what the verdict was. That log is about domains and infrastructure, not about people, and it is what makes it possible to answer a business owner asking why their site was excluded.

Location is asked for, not assumed

Local and map searches need to know roughly where you are, so Marisei asks the browser rather than inferring a profile from your connection. Decline, and you can still search by typing an area name, or by moving the map to the area you care about. Nothing about the location is retained as part of a personal record.

The map is deliberately plain in this respect too: panning and zooming re-runs the search around wherever you settled, so the way you control the area is by moving the map, not by us guessing where you live.

Where AI fits without leaking you

When a search produces a written answer, the model is given the query and material from sources that already passed the trust assessment. It is not given a profile of you, because there is no profile to give it, and the answer is not tuned to who is asking.

Every source cited underneath an answer has been through the same check as any other result, so an answer cannot smuggle in a link that would not have been allowed to appear on its own.

The honest summary

Marisei is less personalised than a mainstream engine and shows fewer results per page. In exchange, there is no history of your searches, no behavioural profile, no advertising auction deciding what you see, and every destination on the page has been assessed before it was offered. If those are the trade-offs you want, the product makes sense. If you want an engine that anticipates you, it never will.