Incognito mode won't tell you your real Google ranking

Why everyone recommends incognito mode to check your Google ranking

Search for how to check your Google ranking and page one will hand you the same answer ten times over: open an incognito window, type your keyword, count down the results until you find yourself. The advice is old, it's free, and its logic feels airtight. If Google shows your own site higher because you visit it daily, then wiping the trail should reveal the neutral result — the one a stranger sees.

The trouble is that the advice was minted around 2010, when Google had just extended personalized search to signed-out users and the filter bubble debate was at its loudest. In that context it made sense: browsing history was the strange, suspicious variable, and erasing it looked like enough to reach a clean result. Fifteen years later the results page barely resembles that list of ten blue links, and history went from being the main source of distortion to being one of the smallest.

There's also a less technical reason the advice survives: it feels good. You open the window, type, count, and walk away with a number. That number feels objective because you saw it with your own eyes. Data from a tool feels borrowed; what appears on your screen feels true. It's the same bias behind every business owner who swears the site is "number one" because searching the brand from the office always puts it at the top. Of course it does — that's a branded query typed from the company's own network.

This isn't an argument that incognito is worthless. It's an argument that it solves a small problem and leaves the big one intact. Below I separate exactly what it clears and what it doesn't, how much each thing weighs, and which conditions a check has to meet before today's number can be compared with next month's. If you'd rather skip to what can be measured continuously and automatically, that's on the Rankiamos features page. If you're here for the argument, keep reading.

What incognito mode does clear (and it's less than you think)

Be fair to the tool before dismantling it. A private window does three real things and does them well. First, it starts with no cookies, so both the site and Google treat you as a first-time arrival. Second, it doesn't write that session into your history, so what you search there won't shape what you see tomorrow. Third — and this is the part that matters most for SEO — it doesn't carry your Google session, so you aren't signed in and account-level signals don't apply.

All of that is fine. The catch is that the list of what it doesn't clear is longer, and its items weigh more. Here's the full split.

Look at the shape of that table. The three green rows are things incognito fixes, and on most commercial or informational queries they move very little. The five amber rows are what's left standing, and that's where both "very high" labels sit. Translated: incognito cleans up the cheap noise and never touches the expensive kind.

There's a fourth point almost nobody makes explicit: incognito doesn't hide you from Google. A private window stops your browser from storing data locally. It doesn't anonymize your connection and it doesn't stop Google inferring where you are. It's a privacy feature aimed at other people who use your computer, not an anonymity mechanism aimed at the search engine. Conflating those two is the root of nearly every misunderstanding on this topic.

Google personalized results: the DuckDuckGo finding that undercuts the advice

So far the argument has been that incognito doesn't touch the variables that matter. Here comes the more uncomfortable part: it doesn't fully solve the one thing it promises to solve either. In December 2018 DuckDuckGo published a filter bubble study in which a group of volunteers searched the same terms at the same time. The relevant conclusion is that most participants saw a different set of results from one another for the same query, and running the search in private mode while logged out barely reduced that variation.

Google publicly disputed the methodology and has argued repeatedly that history-based personalization is limited and concentrated on ambiguous or repeated queries. Both positions can be partly right at once, and you don't need to referee the debate to extract the practical lesson, which is this: even if you side entirely with Google and accept that history-based personalization is small, that weakens the incognito advice rather than saving it. If history barely influences results, clearing it barely changes them. And if it does influence them, the study suggests a private window doesn't fully neutralize it. Either way, incognito isn't the measurement method you were told it was.

What both sides do establish is that Google personalized results are the tip of the iceberg, and a small tip. The bulk of the variation comes from everything above: where you are, in which language, on which device, at which moment, and with which features live on that results page. None of those five is personalization in the strict sense, which is exactly why none of them is fixed by privacy.

Why your Google ranking changes between one search and the next

"Why does my Google ranking keep changing" shows up daily in forums, and the usual answer is one word — volatility — which explains nothing. Breaking it down is more useful, because each cause has a different consequence for how you should be measuring.

1. Cause 1 · The index is always moving

   Google crawls and reprocesses continuously. A competitor's new URL, a changed title elsewhere, a recalculated link — any of those reshuffles the list. It's the most legitimate source of movement and the only one you actually want to measure. Consequence: you need a time series, because only many points let you tell a real move from a bounce.

2. Cause 2 · Google is running experiments

   The results page is permanently under test: formats, blocks, the order features appear in. You can land on one variant and then on another on the next search. Consequence: a single observation proves nothing, and two contradictory observations in a row don't mean something is broken.

3. Cause 3 · SERP features fire and vanish

   The same term can trigger an AI summary in the morning and not in the afternoon, or show a local pack only when Google reads local intent. Consequence: separate organic position from on-screen placement, and record which features were live when you measured. Counting by eye blends the two irreversibly.

4. Cause 4 · You aren't searching the same thing

   "Rank tracker" and "seo rank tracker" are two different queries, as are a singular and a plural, or the same words in a different order. Typing by hand, you rarely enter the identical string two days running, and autocomplete happily nudges you off course. Consequence: the query has to be stored literally, not retyped from memory.

5. Cause 5 · You changed the context without noticing

   Last time you checked from your phone on cellular, this time from the laptop at the office. Or it was 7pm then and it's 9am now. Consequence: context has to be a fixed parameter. That's the line between an experiment and an anecdote.

One frequent confusion deserves clearing up here. People compare what they saw in incognito against the average position figure in Search Console, find they don't match, and conclude one of the two is lying. Neither is: they answer different questions and are computed differently. How that metric is built and why it rarely matches what you see on screen is covered in detail in the piece on Search Console average position versus real position. For our purposes one rule is enough: don't audit an average with a manual search, and don't audit a manual search with an average.

The conclusion of this section is short and it's the hinge of the whole article. Variation is normal and unavoidable. The point of measuring well isn't to eliminate it — you can't — it's to zero out the variables you do control so that whatever movement remains actually means something. That's precisely where manual searching fails: it adds variance instead of removing it.

Enough about what's wrong — here's the standard. A check is decision-grade when today's result is comparable with last week's, and that takes six conditions. They aren't optional or interchangeable: fail one and the rest lose value.

That last row is the decisive one, and it means this debate isn't really about accuracy at all. You can grant that an incognito search gives you a roughly correct number and the argument still holds, because the number isn't stored anywhere. Three months from now, when you want to know whether that keyword improved, you'll have exactly zero comparison points and a vague feeling that "we used to be lower". You can't justify a budget or defend a result to anyone with that.

So the right question isn't "does incognito give me the exact number?" but "does this method build a series?". Manual searching produces loose observations; an automated check produces a series. And everything worth deciding in SEO — whether a change worked, whether a drop is yours or the market's, whether a competitor is eating your ground — comes out of the series, never out of the observation.

One detail on the location point, since it's the objection I get most: "I can force location with DevTools or a VPN." You can. And it works the first time. The problem is reproducing it identically every week, for every city and every device, for months, without slipping once. A method that depends on your discipline to stay intact is already broken.

Building a history from zero: 5 keywords, weekly, free

The practical objection to all of the above is usually financial: "I get the argument, but I'm not paying for a tool to watch three keywords." Fair enough — which is why it's worth grounding the actual starting point, which costs nothing. The Rankiamos Free plan includes 2 projects, 5 keywords, an automatic weekly check, Google Search Console integration and a site scan, with no credit card. The one condition: the weekly check pauses if you go 30 days without opening the app.

5 keywords sounds like nothing until you choose them well. Most small projects don't have five terms that genuinely move the business: they have two or three, plus a long list of variants that change no decision. If you have to pick five, this is the split I use: the head query the money comes from, the variant with the clearest purchase intent, the brand term (so indexing problems surface fast) and two terms sitting between positions 8 and 20, which move the most and teach you the most.

On cadence, be honest about what each rhythm can answer. A weekly check answers "is this keyword trending up or down this quarter?" well. It does not answer "which exact day did it move?", and that second question is the one you need right after shipping a change or when a client asks you to explain a drop. That's what the Pro plan's daily check is for: roughly thirty points a month instead of four, and thirty points is where a step change becomes distinguishable from a bounce. (On Pro, a keyword with 5 or more measurements that goes 30 days without any change drops to a check every 3 days until it moves, and on either plan you can re-measure a keyword by hand 6 hours after its last check.)

Since we're on quotas, the concrete rule: each keyword-by-city combination consumes one unit, so tracking one term across two cities burns two of your 5 Free units. On Free, each keyword is measured on one device, mobile or desktop. On Pro you can pick mobile, desktop or "Both," but each device consumes its own unit: tracking the same keyword on both costs two units, not one. The full per-plan breakdown is on the pricing and plan limits page.

One last point about where the data comes from, because it closes the loop. Queries don't originate from your connection or your browser: they're resolved through DataForSEO using the country and language you set for each keyword, on Google (and also on Bing, on the Pro plan). Your iPhone, iPad, Mac or Apple Watch is the screen where you read the result, not the measurement point. That's where your history, your cookies and your IP stop contaminating the number.