Search Personalisation

Search personalisation is the reason “the number one ranking” is a slight fiction. Two people entering the same query at the same moment can see different results, because Google adjusts what it returns using signals about who and where the searcher is. Understanding which signals matter, and how much, keeps you from chasing a single ranking that does not exist and from misreading your own rank tracking.

Why do two people see different results for the same query?

Google tailors results using a set of signals about the searcher and their context. The main ones, roughly in order of influence:

Location. By far the biggest source of variation between two people’s results, though Google has historically resisted calling it personalisation. Answering the filter-bubble question in 2018, Google’s Search Liaison drew the line at context rather than identity: results “can differ, but usually for non-personalized reasons: location, language settings, platform & the dynamic nature of search”.¹ It is a distinction worth keeping, because the filter-bubble argument below rests on it. Whatever you call it, location does the heavy lifting: results for anything with local intent, and many queries that do not look local, vary by where the searcher is. A search for “plumber” or “coffee” returns nearby options; even “best running shoes” can surface region-specific retailers and prices.

Language and settings. The searcher’s language, region settings, and SafeSearch preferences shape which results are eligible and how they are ordered.

Device. Mobile and desktop results differ, partly through personalisation and partly because layout, local prominence, and page experience weigh differently on each.

Search and browsing history. Google uses recent searches and activity to shape results. Its own description is unambiguous: “with personalization on Search, results you return to often will be promoted in your results when relevant for your search”.² Promotion of sites you revisit is reordering, not refinement.

Account signals. When signed in, broader Google activity can inform results, subject to the user’s privacy settings. Since January 2026 this extends well beyond search history: Personal Intelligence in AI Mode connects Gmail, Google Photos and Calendar as retrieval sources for tailored answers.³ It is opt-in, but it is no longer a niche experiment: as of May 2026 it is available in nearly 200 countries and territories across 98 languages, with no subscription required.⁴

Chosen sources. Users can now nominate sites they want to see more of. Preferred Sources launched for Top Stories in August 2025 and extended to AI Overviews and AI Mode in May 2026. Google reports that “people are twice as likely to click through to a Preferred Source, and people have already selected more than 345,000 unique sources”.⁵

How much does personalisation actually change rankings?

The honest answer changed during 2025 and 2026, and much SEO writing on this topic has not caught up.

The long-standing position, which held for years, was that history-based personalisation was modest: it nudged rather than reordered, and Google actively pushed back on the filter-bubble framing.¹ That statement dates from December 2018, and it is still a fair description of classic ten-blue-links results, where location and language remain the dominant sources of variation between two people.

What has changed is that Google now builds and markets personalisation as a feature rather than defending it as a minor effect. Promoting sites a user returns to often, reading their Gmail and Photos to answer a query, and letting them pin preferred publishers into AI answers are all reordering mechanisms, and two of the three arrived in 2026. The mental model to update: there is still a stable underlying ranking driven by relevance and quality, but the layer on top of it is thicker than it was, and on AI surfaces it is thickest.

Personalisation is also no longer the largest source of variation on AI surfaces. Non-determinism is. SE Ranking ran the same 10,000 keywords through AI Mode three times in one day, from the same locations, on non-logged-in accounts. The average overlap of exact URLs between the three sets was 9.2%, and 21.2% of keywords returned no overlapping URLs at all.⁶ That is the same user, same day, same query, with personalisation held constant. A model that explains result variance purely through personalisation cannot account for it.

What does personalisation mean for rank tracking?

The direct consequence is that a ranking figure is meaningless without its context. Because location and device change results, rank tracking must specify both to be comparable over time:

• Set a location. Track from the location that matches your target audience. A national brand tracks from a representative or neutral location; a local business tracks from its actual service area. An unspecified location produces numbers that drift as the tool’s default changes.
• Separate mobile and desktop. They rank differently; blending them hides real movement.
• Use non-personalised or incognito checks for spot verification. A manual check in a signed-out, history-free session approximates the underlying ranking, but even then location still applies through IP.
• Read your own SERP checks with the same caution. The result you personally see is shaped by your location, device, and history; it is not what your audience sees.
• Sample AI surfaces repeatedly, never once. Given 9.2% URL overlap across three identical runs, a single check of whether you are cited in AI Mode tells you almost nothing. Treat AI citation as a rate measured over many samples, not a position observed once.

Two measurement discontinuities also sit inside the window most readers will be looking at, and neither is personalisation. Google’s removal of the num=100 parameter in September 2025 disrupted how rank trackers collect positions, and a Search Console logging fault inflated impressions between May 2025 and April 2026. Both are covered in rank tracking. The practical rule: read competitive and causal claims on clicks and sessions, which neither artefact touched, rather than on impressions or tracked position.

How should personalisation change your strategy?

It should not send you chasing personalisation signals, because you cannot optimise for an individual’s history and should not try. The productive response is to define your audience and their location clearly, then be the genuinely most relevant, highest-quality result for that audience and place. For anything with local intent, the location signal makes local SEO the priority: a complete and consistent local presence is what wins the location-personalised result.

There is now one exception to “not a lever to pull”. Preferred Sources is user-chosen personalisation a publisher can actively ask for. A reader who adds you sees more of your content in Top Stories, AI Overviews and AI Mode, and Google reports Preferred Source links attract roughly double the click-through rate. Any site is eligible if it publishes fresh content. Google explicitly notes that publishers are encouraging their readers to use it,⁵ so prompting your audience to add you is a legitimate and unusually direct action, and one of the few places where personalisation can be influenced rather than only accounted for.

Everywhere else the durable advantage is relevance and quality, which survive personalisation. There is no shortcut through the signals themselves.

Footnotes

1. Google says their personalised results do not create filter bubbles (Danny Sullivan, Search Liaison), December 2018 — Search Engine Roundtable ↩ ↩²

2. 6 ways to customize Search for more relevant results — Google ↩

3. Personal Intelligence in AI Mode in Search — Google ↩

4. Google Search’s I/O 2026 updates: AI agents and more — Google ↩

5. New ways to find your favorite sources and original content in AI Search — Google ↩ ↩²

6. AI Mode research: what 10,000 keywords reveal — SE Ranking ↩