By Mark Buraga, Independent SEO Consultant, Growth Engine PH. Last updated 14 June 2026.
Your rankings can hold steady while your traffic quietly falls. That is not a tracking glitch. The result people used to click moved into an AI answer, and the page that earns that answer is often not the one sitting at the top of the blue links. Position stopped telling the whole story.
So tracking keyword rankings no longer means watching one number between 1 and 100. The job became three layers: position (where you rank), presence (whether you are cited or mentioned in the AI answer), and share of voice (whether you are gaining ground across a set of queries, not moving on one term). Track all three and you measure reality. Track only the first and you are optimizing a number that already decoupled from the outcome. Only 38 percent of pages cited in Google AI Overviews also rank in the organic top 10, down from roughly 76 percent eight months earlier (Ahrefs), and 68 percent of US Google searches now end without a click (SparkToro). The layer your old report watched and the layer that wins traffic are no longer the same thing.
This is a practical, tool-agnostic guide to tracking all three layers: the signal-versus-noise rule, what Google Search Console actually measures, and a simple routine that ties it together. It builds on how rank reporting became multi-surface visibility rather than a single ranking column.
What Tracking Keyword Rankings Means Now (and Why Position Is Only Part of It)
Tracking keyword rankings now means tracking three layers: position (where you rank), presence (whether you are cited or mentioned in the AI answer), and share of voice (whether you are gaining ground across a set of queries).
The single-number model broke for a simple reason. The answer people used to click moved up into the AI Overview, and the pages cited in that answer are mostly not the pages ranking at the top of the organic results. The decoupling is observable, not asserted: across a study of 863,000 keyword results and four million AI Overview URLs, only 38 percent of AI-Overview-cited pages also rank in the top 10 (Ahrefs), and a separate Ahrefs cut puts the overlap at about 12 percent for the original prompt. When the page that earns the answer usually is not the top-ranked page, a tracker that only watches the top 10 is watching the layer that came loose from your traffic.
The other half of the story is the click. In the first four months of 2026, 68.01 percent of US Google searches ended without a click, up from 60.45 percent in 2024; for every 1,000 US searches, only 276 clicks now reach the open web (SparkToro). So a ranking is no longer the same question as traffic. Position one on a query an AI Overview answers in full can send fewer clicks than position five on a query it does not touch.
None of this means position is dead. It means position is layer one of three, and on its own it has stopped predicting the result. The fix is not a better rank tracker; it is a wider definition of what tracking means. The rest of this guide is how to run each layer.
Layer One: Track Position Accurately (and Stop Reacting to Noise)
Act on a sustained shift of five or more positions over 7 to 14 days, and ignore a one-to-three position daily wobble, because that is personalization noise, not a ranking change.
Position is still worth tracking. The problem is that raw position data is noisy, and most people react to the noise instead of the signal. Google personalizes results by location, device, and search history, so the position you see in your own browser is not the position a tracker reports, and two tools will often disagree on the same keyword by a place or two. That is why incognito checks in your own browser are unreliable.
A dedicated rank tracker earns its keep by stripping personalization to a neutral baseline: it checks from a consistent location and device profile, signed out, so the number is comparable week over week. Once you have a clean baseline, act only on signal. A keyword that moves from position 8 to position 3 and holds there for two weeks is worth investigating. One that bounces between 4 and 6 every day is noise, and chasing it wastes time you should spend on the work that actually moves rankings.
This is the honest answer to the common worry about whether rank checkers are accurate at all. They are accurate enough when you read them correctly: trust the trend, not the daily reading, and set your threshold at a sustained five-position shift. If you want depth on baselines and cadence, this rank-tracking guide is a solid reference. The method matters more than the tool; any tracker that holds a neutral baseline and shows a trend line will do.
Why Google Search Console Average Position Is Not Your Ranking
GSC average position is a blended average across queries and pages, inflated by branded terms; it is essential for clicks and CTR, but it is not a per-keyword ranking.
This is the most misunderstood number in the whole stack, so it is worth being precise. Google Search Console average position is one of the most misunderstood performance metrics in SEO (SEOTesting). It is not where you rank for a keyword. It is an average of every position your pages held across every query they appeared for, weighted by impressions. Your branded searches, where you rank first almost by default, pull that average up and make it look healthier than your real visibility on the queries that win new work.
What GSC is excellent at is the click side: it is your source of truth for clicks, impressions, click-through rate, and the actual queries people used to reach you. That query-level data is something no rank tracker gives you, because only Google knows what people typed. But GSC does not tell you where you rank for a specific term on a neutral baseline, and it does not tell you whether an AI answer cited you.
So the practical split: use GSC for clicks, CTR, and query discovery; use a rank tracker for clean per-keyword position; and use a separate check, covered next, for AI-answer presence. Treating average position as a ranking is how people convince themselves everything is fine while the queries that matter quietly slip. It is one input to the stack, not the stack.
Layer Two: Track Presence in AI Answers (the Layer Most Guides Skip)
Presence tracking asks whether AI answers cite or mention you when someone asks a question in your category, and it is now the layer that best predicts whether you get a click.
This is the layer almost every guide published this year still skips, and it is the one the decoupling data points straight at. If the page that earns the answer is usually not the top-ranked page, then watching position alone leaves you blind to whether you are in the answer at all. Presence closes that gap. It is the AI equivalent of rank tracking for classic search, except the results are conversational, not a list of blue links (Averi).
The first time I ran my own site through the full three-layer check instead of a positions report, the gap was immediate. On the branded queries I was everywhere: ranked, cited, mentioned. On the non-branded discovery queries that actually win work, the position column looked unremarkable but not alarming, a few things on page two, nothing on fire. Then I looked at the presence layer. On several of those discovery queries an AI Overview was forming the answer, and I was not in it. Not ranked low in it; absent entirely. A positions-only report would have shown a slow, boring climb and nothing to worry about. The presence layer showed the real problem: the answer was being assembled without me, on the exact terms a buyer would use. The number that mattered was not where I sat in the list. It was whether I was in the answer at all, and I was not.
You do not need to buy anything to start tracking this. The free manual method is straightforward:
- Write a fixed prompt set, ten to twenty real questions a buyer in your category would ask, in plain language.
- Run that same set across multiple engines, Google AI Overviews, ChatGPT, Perplexity, and Gemini, on a fixed cadence.
- For each prompt, log whether you were cited as a source, mentioned in the answer body, or absent. Keep the prompts and the cadence identical so the comparison holds over time.
Run it across more than one engine, because the engines barely agree with each other. Only about 11 percent of domains are cited by both ChatGPT and Perplexity across an analysis of 680 million citations (Averi), a figure a separate 118,000-response study confirms. Tracking one engine misses most of where you are, or are not, cited. A single check is not presence tracking; a fixed multi-engine prompt set on a cadence is.
Presence is the measurement. Earning it is a different job, and it routes elsewhere. When you find you are absent, start with how AI engines choose sources and the underlying citations versus mentions distinction, then audit why a site is not cited. This page is about tracking the outcome. Those explain how to change it.
Not sure how to build a prompt set for your category? That is exactly the first thing an AI-visibility read sets up.
Layer Three: Track Share of Voice, Not Just Single Keywords
Share of voice estimates your expected clicks across a whole keyword set, so it shows whether you are gaining market share instead of just moving on a handful of terms.
Most guides still teach you to pick your keywords and watch them climb. The trouble is that a handful of keywords moving up tells you almost nothing about whether you are winning. You can move three terms into the top 10 and lose ground overall if the queries that drive real volume slipped at the same time. Share of voice fixes the framing by measuring the set, not the term. It measures how much impact you create, whereas keyword rankings show where you stand (Keyword.com), and it is a truer signal than saying you have 20 keywords in the top 10.
Mechanically, it weights each keyword by its search volume and your expected click-through at the position you hold, then expresses your slice of the total available clicks for that set as a percentage. The number matters less than its direction. Rising share with flat positions means you are gaining ground where the volume is. Falling share while a few vanity terms climb means you are losing the queries that count.
To make it work, define the set deliberately: pick the queries that represent real buying intent in your market, exclude your branded terms so you are not flattering yourself, and hold the set steady so the trend means something. A clean keyword set tracked as one share-of-voice figure is the closest thing rank tracking has to a market-share number, and it answers a question single positions never could: am I gaining, holding, or slipping across the whole front.
If your current report is all positions and no presence, it is measuring the layer that decoupled. Ask for a measurement audit.
How Often Should You Check Rankings?
Check position weekly, presence in AI answers weekly to monthly across a fixed prompt set, and share of voice monthly; daily checking trains you to react to noise.
Cadence is where good intentions go wrong. The instinct after a content push is to refresh the rank tracker every morning, but daily checking is the fastest way to mistake a one-to-three position wobble for a trend and start tinkering with pages that were fine. Match the cadence to how fast each layer actually changes, and the noise mostly solves itself.
| Layer | Check cadence | Why this cadence |
|---|---|---|
| Position | Weekly | Daily readings are mostly personalization noise; a week is long enough to see real movement |
| Presence in AI answers | Weekly to monthly | AI answers shift more slowly than rankings; a fixed prompt set run on a steady rhythm shows the trend |
| Share of voice | Monthly | A set-level market-share figure is a trend metric, not a daily reading |
| Clicks, impressions, CTR (GSC) | Weekly | The outcome layer; watch it alongside position to catch ranking-holds-but-traffic-falls early |
The cadence ties directly to the signal-versus-noise rule from layer one. You check weekly precisely so you can apply the five-position, 7-to-14-day threshold before you act. Daily checking does not give you better information; it gives you more chances to react to movement that was never real.
Free vs Paid: Building a Tracking Stack on Any Budget
You can run all three layers for free to start: GSC for clicks and CTR, a free rank checker for position, and a manual prompt set for AI-answer presence; paid tools mainly buy you scale and automation.
You do not need to spend anything to begin tracking all three layers properly. Here is what each budget level buys, by category, not by brand.
Free to start. Google Search Console covers clicks, impressions, CTR, and query data at no cost, and it is non-negotiable regardless of budget. A free rank checker covers per-keyword position for a small set of terms. The manual prompt-set method covers presence, and a simple spreadsheet covers share of voice if you are willing to do the weighting yourself. That is a complete three-layer stack for zero spend, and the right place to start before you buy anything.
Paid, when you outgrow the manual version. What money buys is scale and automation, not a different method. A dedicated rank tracker watches hundreds of keywords on a clean baseline and stores the history. An AI-visibility tracker runs your prompt set across engines on a schedule, the same thing the manual method does, faster and at volume. Share-of-voice reporting becomes a chart instead of a spreadsheet formula. None of it changes what you are measuring; it removes the manual labor once your keyword set is too large to log by hand.
The decision rule is simple: start free, prove the method on a small set, and upgrade a layer to a paid tool only when the manual version stops scaling. Pick tools by the layer you need to automate, not by the longest feature list. The category matters; the specific logo is yours to choose.
Putting the Three Layers Together: A Simple Tracking Routine
A complete routine logs position, presence, and share of voice on a set cadence, acts only on sustained signal, and treats AI-answer presence as a first-class metric, not an afterthought.
Here is the repeatable routine, start to finish:
- Weekly: pull position from your tracker on a neutral baseline, and pull clicks, impressions, and CTR from Google Search Console. Flag any keyword with a sustained five-position shift over the trailing 7 to 14 days. Ignore the daily wobble.
- Weekly to monthly: run your fixed prompt set across Google AI Overviews, ChatGPT, Perplexity, and Gemini. Log cited, mentioned, or absent for each prompt on each engine. Watch the trend, not a single run.
- Monthly: calculate share of voice across your defined keyword set and note the direction. Rising share with flat positions is good news; falling share with a few climbing vanity terms is the warning.
- Always: act on signal, not noise, and read GSC average position as a blended average, never as a ranking.
The point is not the logging. It is that the three layers together catch problems a positions-only report hides. When rankings hold but traffic falls, the presence layer usually shows why. When a few keywords climb but the business is not growing, share of voice usually shows why. The single number never could.
This is the work the AI era actually rewards, and it is how I run measurement for the businesses I work with at Growth Engine PH. As an independent SEO consultant, I would rather hand you a method you can verify yourself than a dashboard you have to take on faith.
Want a read on whether AI answers cite you on the queries that win you work, not just where you rank? Tell me your site and your market, and I will run the three-layer check. Talk to Mark.
Position tells you where you sit. Presence tells you whether you are in the answer. Track both, or you are guessing.
Frequently Asked Questions
How do I check my SEO rankings? Use Google Search Console for clicks, impressions, and CTR, plus a rank tracker for a neutral per-keyword position. Do not rely on manual incognito checks in your own browser, because Google personalizes results by location, device, and history, so what you see is not what a clean baseline reports (SEOTesting).
How accurate are keyword rank checkers? Accurate enough when you read them correctly. Trackers strip personalization to a neutral baseline, but positions still wobble day to day, so act on a sustained shift of five or more positions over 7 to 14 days rather than reacting to daily noise. Trust the trend, not the single reading.
Is there a free website keyword ranking checker? Yes. Several free checkers cover per-keyword position, and Google Search Console covers clicks and CTR at no cost. You can run the full three-layer stack for free to start: GSC for clicks, a free checker for position, and a manual prompt set for AI-answer presence.
Is keyword ranking a KPI? Position is a leading indicator, not the outcome. With 68 percent of US searches now ending without a click (SparkToro), the KPIs that map to traffic are clicks, share of voice, and AI-answer presence. Track position, but do not report it as the result.
Why am I ranking but getting no traffic? Because zero-click results and AI Overviews intercept the answer before the click. Only 38 percent of AI-Overview-cited pages also rank in the top 10 (Ahrefs), so ranking no longer guarantees a click. That gap is exactly why presence in AI answers is now a tracked layer, not an afterthought.
How do I track if my site is cited in AI Overviews or ChatGPT? Run a fixed prompt set of real category questions across multiple engines on a set cadence, and log for each prompt whether you were cited, mentioned, or absent. Track more than one engine, because only about 11 percent of domains are cited by both ChatGPT and Perplexity (Averi); a single-engine check misses most of the picture.
Should I use Google Search Console or a rank tracker? Both. Use GSC for clicks, CTR, and the queries people actually used; use a rank tracker for a clean per-keyword position on a neutral baseline and for share of voice. Neither one, on its own, tells you whether an AI answer cites you, which is why presence is a separate layer (SEOTesting).
How often should I check my keyword rankings? Weekly for position, weekly to monthly for AI-answer presence across a fixed prompt set, and monthly for share of voice. Daily checking mostly trains you to react to personalization noise instead of real movement (Nightwatch).
Work with us: explore Growth Engine PH SEO services, built to earn AI citations and rankings.
Related reading: What is SERP optimization and the 6-Block GEO Diagnostic.