Incognito Is Not a Neutral Search
A private window removes some personalisation and none of your geography. Your IP address still says where you are, so the result page is still localised to your location — and location is the largest single source of variation in most result sets. You’ve removed the smallest input and kept the biggest one.
This matters because hand-checking in a private window is how most people form their intuition about where they rank, and that intuition then argues with the tracker.
What a private window does and doesn’t drop
Dropped: your signed-in state, your search history in that session, and cookies from previous sessions. Genuinely useful — signed-in history is a real personalisation input, and repeatedly clicking your own result can influence what you see afterwards.
Kept: your IP address, and with it your approximate location. Also your device type, your operating system and browser, your interface language, and your network’s general characteristics.
So a private window answers “what does someone in my city, on my device, with no history, see right now?” That’s a legitimate question. It is not “where do we rank”, and it’s not the question your tracker is answering unless you configured the tracker to match your own city and device, which almost nobody does.
Why the mismatch causes arguments
The recurring scene: someone checks a keyword in a private window, sees position three, and asks why the tracker says eight. Both numbers are right. The tracker sampled a national location; the browser sampled the office, which happens to be near the business, on a query with local intent.
The unhelpful resolutions are to distrust the tracker or to reconfigure it to agree with the browser. The useful resolution is to state which question each number answers and then decide which one the report needs — which is almost always the tracker’s, because it’s reproducible and comparable over time.
The underlying point is the same one behind why two rank trackers disagree: the request determines the answer, and consistency beats being occasionally exactly right.
Other ways manual checking misleads
Repeated checking of your own keywords. Even without history, you’re a real session with real behaviour, and you are the least representative searcher for your own site.
Search suggestions and autocomplete. Typing a partial query and taking a suggestion changes the query. People do this constantly and then report on a term they didn’t intend to check.
Scroll-and-count errors. Counting organic positions by eye on a feature-heavy page is genuinely hard, and it’s the same ambiguity trackers resolve with explicit rules — see what position one means on a crowded result page.
No record. A manual check produces a number and no history, so it can’t tell you whether anything is unusual — you can’t read a position without a baseline.
When manual checking is the right tool
It has real uses, all of them qualitative:
- Looking at the page. What blocks are present, who’s above you, what the intent of the page appears to be. A tracker gives you a number; your eyes give you the layout.
- Sanity-checking a suspicious tracker reading. If the tracker says a keyword vanished, opening the page tells you quickly whether something structural happened.
- Checking what a customer in a specific place sees, if you happen to be in that place.
- Confirming a page is being shown at all after a change.
In every case the output is an observation, not a position for the record.
Getting closer to neutral
If you must hand-check, at least be deliberate. Use a private window to drop history, set the interface language explicitly, and be aware your IP is fixing the location. Use a device that matches the device you care about. Write down where you were.
Better: use the tracker for the number and the browser for the picture. That division of labour removes the whole argument, and it’s the same split as rank tracker vs. Search Console at a smaller scale.
What to actually do
- Stop using hand-checked positions as data. They go in no report.
- Use manual checks to read the page, not to read your rank.
- When a manual check disagrees with the tracker, name the two questions rather than picking a winner.
- Note your own location when you hand-check, because that’s the variable doing the work.
- Never reconfigure a tracker to match your browser. You’d be tuning a reproducible measurement to agree with an unrepeatable one.