How Should You Track Rankings When Everyone Sees a Different Result?

How Should You Track Rankings When Everyone Sees a Different Result?

Stop treating a rank as a fact about your page and start treating it as an average across many different searches. Google itself reports position as an average, not a single number, and warns that the number moves with device, screen width and the type of result. A rank tracker showing one clean number is hiding that spread.

We get asked about this a lot. A founder opens a rank tracking tool, sees position four, then searches the term on their phone and finds their page nowhere. Both things are true at once.

This piece is our argument for what to measure instead, and where third party rank tracking still earns its cost.

Why Does the Same Search Give Two People Different Results?

Because position depends on things that are not your page. Google's Search Console Help documentation is direct about it. The docs say that positions vary significantly by device, screen width and result type, and that a user's search history and location affect individual result positions differently than the averages you see in reports.

Location is the biggest one for most B2B companies. A search made in London and the same search made in Sydney are two different searches to Google, even with identical words.

Then there is the layout. A result that sits above the fold on a wide desktop monitor can sit below three other blocks on a phone. Google counts position top to bottom, so a page that gained a block above it lost position without changing at all.

What Does Google's Average Position Actually Mean?

It means the topmost position your site reached, averaged across every query where you appeared. Google's documentation gives the worked example. If a query returns your site at positions 2, 4 and 6, that counts as position 2. If another query returns you at 3, 5 and 9, that counts as 3. The reported average is 2.5.

Two things follow from that, and both surprise people. First, the number is an average of averages, so a single new query you barely rank for can drag it down while nothing about your existing pages changed. Second, appearing three times on one page does not help the number. Only your best placement counts.

This is why we tell clients that a moving average position is often a story about which queries they became visible for, not about whether their pages got better or worse.

How Do AI Overviews and AI Mode Change the Count?

They are counted using the same impression and position rules as anything else. Google's documentation states that an AI Overview occupies a single position, and every link inside it is assigned that same position. A follow up question inside AI Mode is treated as a new query with its own metrics.

That has a practical consequence. Being cited inside an AI Overview at the top of the page can register as a strong position even though the user may never click, which is the tension at the centre of building a strategy for zero-click search.

Google also describes how these results are assembled. Its documentation on AI features says the system uses query fan-out, issuing multiple related searches across subtopics to build a response. So the query you track is not necessarily the query that surfaced your page.

Google is equally clear on the other side of that. Its documentation states there are no additional requirements to appear in AI Overviews or AI Mode, and no special optimizations necessary. There is no separate rank to chase.

Is Third Party Rank Tracking Still Worth Paying For?

Yes, but for competitive intelligence rather than for grading yourself. Tools like Semrush, Ahrefs and Sistrix run their own searches from chosen locations and devices. That is a controlled sample, which makes it useful for comparison even when it is a poor description of reality.

The honest framing is that a rank tracker measures a robot's search, not your buyer's search. That is a feature when you want a stable yardstick between you and a competitor. It is a bug when you report it as what customers see.

We think the failure mode is not using the tools. It is reporting the tool's number to a board as though it were the truth, then having to explain a drop that was really a layout change.

What Should You Measure Instead?

Measure impressions and clicks by query group, and watch the direction over weeks rather than days. Search Console gives you real data from real searches rather than a simulated one. Clicks divided by impressions gives you click-through rate, which tells you whether your listing is earning the attention your position should be getting.

Group your queries before you look. Branded searches behave nothing like problem-aware searches, and averaging them together produces a number that describes neither. We usually split into brand, category, competitor and problem queries.

Then look for the pattern that matters most: impressions up and clicks flat. That means Google is showing you more often and people are choosing something else. That is a title, description and page problem, not a ranking problem.

This is the same discipline we apply when helping teams decide what to do with marketing attribution in B2B SaaS. Pick the measure that survives contact with how the system actually works.

How Often Should You Actually Look?

Monthly for reporting, weekly at most for diagnosis. Search Console data is noisy day to day, and reacting to a two day dip is how teams end up rewriting pages that were fine.

There is one exception. After a migration, a redesign or a big content change, check daily for two weeks. You are not tracking rank then. You are watching for a cliff that means something broke.

Outside those windows, daily rank checking is mostly an anxiety habit. We say that as people who have had the habit.

What About Tracking Visibility in AI Answers?

Treat it as a separate measurement problem with weaker tools. There is no official position report for ChatGPT or Perplexity. The practical approach is to run a fixed set of real buyer questions through the assistants your customers use, on a schedule, and record whether you were named.

Keep the question list small and stable. Twenty questions checked every month will tell you more than two hundred checked once.

Record the answer text, not just a yes or no. What the model says about you is usually more actionable than whether it mentioned you, and it is the fastest way to find out that something outdated is being repeated about your product.

What Should a Monthly Search Report Contain?

Four things, in our view. Clicks and impressions split by query group. Click-through rate on your top twenty queries. A short list of queries where impressions grew but clicks did not. And a note on what changed on the site that month.

Leave average position in the report, but put it last and label it as an average. It is context, not a score.

The fourth item is the one most reports skip and the one that makes the others readable. Without it, every movement looks like weather.

Where Does This Leave the Idea of Ranking Number One?

It leaves it as a useful goal and a bad metric. Being the best answer to a question still drives everything. It is just that the scoreboard for it was never as precise as the tools implied, and the gap between the number and reality keeps widening as results get more personal and more assembled.

The teams we see doing this well have made a quiet switch. They stopped asking where do we rank and started asking which questions are we visibly answering, and are people choosing us when we appear.

If you want help rebuilding your search reporting around questions rather than positions, or a website that can actually answer them, we are happy to talk it through. You can reach us at phoenix.studio.