Rank tracking looks simple: pick keywords, check positions, draw a line. Then the numbers disagree with each other, jump around for no reason, and stop explaining traffic once an AI Overview sits above the results. Most rank reports mislead the people who read them, not through bad data but through reading it the wrong way.
This post covers the two sources of ranking data, how to read each correctly, how to account for AI Overviews, and how to report rankings so they lead to decisions.
Where does ranking data come from?
Two places, and they measure different things.
Google Search Console
- What it measures
- Google's own record of your impressions, clicks and average position
- Strength
- Real data from every searcher
- Limit
- Only your site; no competitors; no view of the page
Rank tracker
- What it measures
- A scheduled search from one location and device, recording the results page
- Strength
- Shows competitors, features and the full page
- Limit
- One sample, not every searcher
Use both. Search Console tells you what happened. A rank tracker tells you what the page looks like and who you are up against.
How do you read Search Console's average position?
Google defines average position as "the average position of the topmost result from your site". Three things follow from that definition, and each catches people out:
- It is averaged across every impression. Different searchers, locations and devices see different positions. A query where you are third for some people and twelfth for others averages somewhere in between.
- It counts your topmost result only. If two of your pages appear, only the higher one counts.
- It only counts when you appeared. A query where you rarely appear, but rank well when you do, can show a flattering average with few impressions.
So always read average position beside impressions and clicks. "Position 4.2 on 30 impressions" is a different fact from "position 4.2 on 30,000 impressions". Google defines impressions as "how many times your site appeared in Search results" and clicks as "the number of times a user clicked your site from Google Search results".
The Performance report shows the past three months by default. If you want history longer than Google keeps, export it regularly or use a tool that stores it day by day.
How do you set up a rank tracker sensibly?
Choose the searches carefully
Start from your own Search Console data: the queries that already bring you impressions are proof that Google associates you with them. Add the buying searches you want to win but do not appear for yet. Fifty to a hundred searches is plenty for most businesses.
Match the location and device to your buyers
Rankings differ by country, city and device. Track from where your buyers search. If you sell in two countries, track both separately.
Record more than your position
Record the top ten results, so you know who outranks you; your ranking URL, so you notice when Google swaps which of your pages it shows; and whether an AI Overview, featured snippet or other feature appeared.
Treat failed checks as unknown
A check that failed or was blocked is not "not ranking". Store it as missing, and leave it out of averages.
The table at the top is a worked example of five such readings for one search, on an example site, not a real reading. Each day keeps the position (or why there is none), the page that held it, and the AI Overview as true, false or unknown. The 3 September check failed, so it has no position and an unknown AI Overview, and an average leaves it out; on 5 September the site was outside the first 30, which is a measured result, not a missing one.
Why do rankings jump around?
Because a single check is a single sample. Results vary between checks through normal reshuffling, personalisation, data centre differences and tests Google runs. A query that moves from 5 to 8 and back to 5 over three days has not changed.
Read weekly averages for decisions and monthly trends for reports. React fast only to a large drop that persists for several days on an important query, and check it is not a tracking failure before you react.
How do AI Overviews change rank tracking?
An AI Overview sits above the organic results and can answer the query before anyone scrolls. It cites its own set of pages, which may or may not include the pages ranking first. Your position can stay the same while your clicks fall, because the answer moved above you.
So for each tracked search, record three things about the AI Overview:
- Was one shown? Yes, no, or unknown when the check could not tell. Do not count unknown as no.
- Which pages did it cite?
- Were any of them yours?
Then read your position, your clicks and the AI Overview together. A stable position with falling clicks and a new AI Overview that does not cite you is a different problem from a falling position.
Google says a page needs to be indexed and eligible for a snippet to appear in AI Overviews, with no special markup. The work to be cited there is the same work as for featured snippets: see how to optimize for featured snippets.
What should a ranking report contain?
One row per important search, with:
| Column | Why |
|---|---|
| Search | The query, in the buyer's words |
| Average position (Search Console) | Google's own view |
| Impressions and clicks | Whether the position matters |
| Tracked position (rank tracker) | The page as a searcher sees it |
| Top competitor above you | Who to study |
| AI Overview shown, and cites you? | Whether position still means clicks |
| Window | The dates the numbers cover |
Below the table, three lines: what moved, what we think caused it, and what we will do. If nothing moved beyond normal noise, say so. "No change we can prove" is a valid finding.
What about rankings in AI answers?
Google rankings are one surface. Your buyers also ask ChatGPT, Perplexity, Claude and Gemini, where there are no positions in the same sense: an answer names some brands and cites some pages. The equivalent of rank tracking there is counting how often you are named, out of how many answers, per engine, and which pages are cited. See how to calculate share of voice and how to audit your AI mentions.
Keep the two side by side. A search where you rank first on Google and are never named in AI answers tells you the AI engines are reading different sources than Google's ranking rewards.
