Rank tracking: the number you see is not the number anyone else sees

The accuracy percentage nobody can check

Search "rank checker accuracy" and you will find claims like a specific tool hitting 96.86% accuracy, or top-tier tools landing "between 95% and 100%," repeated across dozens of 2026 posts with no methodology, no sample size, and no test date attached to any of them. A percentage with no visible test behind it is marketing copy wearing a lab coat.

Ahrefs' own help documentation is more honest about the actual mechanics, and worth reading precisely because it is a vendor admitting the limits of its own product rather than selling past them. It names four concrete reasons a tracker's reported position will not match what you see: results below the top five fluctuate more and can shift by several positions between the tool's crawl and your manual check; a time lag of hours to days sits between when the tool pulls data and when you look; Google adjusts results by location, so a colleague across town can see a different order than you do; and Google customizes results based on your own search history and account activity when you are signed in. Its fix is the one any rank tracker has to use to mean anything at all: check in an incognito window while signed out, and if you want to compare cities, use a VPN or the tool's location targeting rather than your own browser.

Two studies, thirteen years apart, that disagree

How much personalization actually changes is the harder question. The two most rigorous public attempts to measure it landed in different places.

A 2013 analysis of Google search data from Northeastern University researchers, summarized by Briggsby, found personalization altered 11.7% of results on average, comparing logged-in users and varying IP addresses against a control. The effect was not even across query types. Political queries showed the most movement, at 57%, while plain factual "what is" queries showed the least, at 18%. When the top result did change because of personalization, 73% of the time the replacement was already sitting at position two, and 93% of the time it came from somewhere on page one.

A December 2019 peer-reviewed study in EPJ Data Science took a different approach entirely. Researchers collected search results from more than 4,000 volunteers during Germany's 2017 federal election, comparing what different users saw for the same political and candidate-name queries over 27 weekdays. Their conclusion ran against what the personalization narrative usually assumes: "the room for personalization of the search results is very small," with fewer than four in ten results varying for party searches and fewer than two in ten for candidate searches. What looked like personalization was driven mostly by geographic regionalization rather than an individual's browsing profile, and the paper describes the resulting filter-bubble effect as negligible.

Neither study has been repeated since, on English-language commercial queries, at meaningful scale. One says personalization moves more than a tenth of results and can swing well past half for sensitive topics. The other, using tighter peer-reviewed methodology six years later, says the room for it is small and geography does most of the work anyone mistakes for a personal profile. Both cannot be fully right. Nobody has re-run either test since AI Overviews existed, let alone since 2026's new personalization tools shipped. That gap is the honest starting point for anything a rank tracker claims to measure today.

Then Google added a fifth input in May 2026

Whatever the 2013 and 2019 numbers say, the input side of the equation has grown since. Google's own Search blog announced on 19 May 2026 that Personal Intelligence, which connects a signed-in user's Gmail and Google Photos, with Calendar coming soon, had expanded to "nearly 200 countries and territories across 98 languages," no subscription required, framed as a step toward AI responses shaped by a user's own context rather than the same answer for everyone.

Search VP Liz Reid described a related, more direct lever in March 2026: a preferred-sources feature where, in her words, if a user indicates they trust a given site, "Google will show that site more often." That is personalization built by intent rather than inferred from behavior, and it comes from the person running Search, not a theory about what the algorithm might be doing.

Neither Google announcement includes a measurement of how much either feature moves an average commercial ranking, and no independent researcher has published one yet. They describe reach and mechanism, not magnitude. The honest position in September 2026 is that Google has added at least one confirmed new personalization input this year, on top of the location and login-state factors already established, and nobody outside Google can currently say how much it moves a typical keyword's position for a typical user.

What a tracker can still promise

None of this means rank tracking is pointless, and it does not mean your tracker is lying to you. It means a single manual search, on your own laptop, in your own city, while signed into your own Google account, is one personalized instance of the results, not a neutral reference point the tracker failed to match. Our companion piece on whether rank tracking still matters covers the separate problem of AI Overviews collapsing click-through even when a position holds. This is the layer underneath that: the reason the position itself is not one fixed fact to begin with.

What a tracker can honestly deliver is a fixed, disclosed methodology, checked from the same signed-out state, the same declared location, and the same schedule every time, so the number that moves week to week is comparable to itself, even though it was never going to match what any one signed-in person happens to see on any one day. Local rankings take this problem further, since a business's map pack position can change block by block depending on exactly where the searcher stands, the subject of our guide to local SEO and AI search. And when a client says "I searched it and my competitor is above me, your tool is wrong," the first question worth asking is what device, location, and login state they used, a confusion detailed further in our guide to diagnosing why a competitor outranks you, before treating one screenshot as ground truth.

What we could not verify

The 2013 Northeastern figures predate mobile-first indexing, BERT, AI Overviews, and Personal Intelligence by years. Whether an 11.7% average personalization rate holds on today's search results has not been retested. The December 2019 EPJ Data Science study covered only German-language political queries during one election period, a category its own authors note is more likely to show personalization than an average commercial search, and it has not been replicated on English-language or transactional queries. Neither Google's Personal Intelligence announcement nor Liz Reid's remarks on preferred sources come with a published figure for how much either feature shifts organic position, so this piece describes what each does, not how much it moves rankings, because nobody has published that number yet. Accuracy percentages like the 96.86% figure circulating in 2026 rank-tracker marketing carry no visible sample size, test date, or methodology, so we could not verify them and have not repeated them as fact.

A check worth running this week

1. Search one of your tracked keywords yourself, once in a private or incognito window with location sharing off and signed out of Google, and compare that position to what your tracker reported for the same day.
2. Run the same search again from a different device or network and note whether the position changes. That gap is your own live demonstration of the problem this piece describes, not a tool malfunction.
3. Before treating a client's or competitor's "I searched it and we're not there" as evidence your tracking is broken, ask what device, location, and login state produced that one screenshot.

Sources

The methodology explanation is from Ahrefs, Why don't the ranking positions in my Rank Tracker reports match those I see in Google?, Ahrefs Help Center. The 2013 personalization figures are from Briggsby, A Better Understanding of Personalized Search, 24 June 2013. The 2019 study is Cornelius Puschmann et al., What did you see? A study to measure personalization in Google's search engine, EPJ Data Science, 16 December 2019. The Personal Intelligence expansion is from Google, Google Search's I/O 2026 updates: AI agents and more, 19 May 2026. Liz Reid's remarks on preferred sources are reported in Google's Liz Reid On Progression Of Google AI Search, Search Engine Roundtable, 9 March 2026.

Common questions