The most dangerous failure a rank tracker can have is not an error message. It is the failure that looks exactly like a normal day: the graph updates, the timestamps move, and the number quietly says "not in the top results" — when in reality the tracking pipeline, not your rankings, is what broke. If you cannot tell "nothing happened" from "it stopped working", every drop alert you have ever received is suspect, including the ones that were real.
We run a rank tracker ourselves, and in late October 2026 we spent six days inside exactly this problem. This post is the checklist we wish we had beforehand, plus what the incident taught us about which failures a tracker can hide from its own user.
The Three Silent Failure Modes
Everything below reduces to one of three shapes:
- Fabricated numbers. The tracker writes a position it did not actually verify — from a stale cache, a partially readable result page, or a snapshot taken under different conditions. This one is worst, because it generates confident alerts about movements that never happened.
- Stalled collection. Checking stops — quota exhausted, function killed by a timeout, API contract changed — but the dashboard keeps showing the last good number as if it were current. Nothing looks wrong until you notice the date under the graph.
- Manufactured absences. The check runs, the API answers, but the payload is unusable — and the tool counts it as "your site is not in the top results". No error, no warning, and a false drop lands in your history permanently.
The third one deserves its own section, because in late 2026 it stopped being a rare glitch.
The Poisoned-SERP Incident, From the Inside
Since September 2026, Google has been serving ranking tools search results that look complete and normal but have the actual results swapped out. Barry Schwartz documented the pattern: roughly 8% of tracked SERPs were affected, the swap passes within days, and nothing on the page itself reveals it — you can only detect it against a keyword's own history.
We got six consecutive days of it starting October 31. On our worst checks, most of the "organic results" in the payload carried unresolvable redirect links instead of real domains. The target domain was not invisible because it lost rankings — it was invisible because the data itself was fake. A tracker that parses that payload and trusts it will report "not found" for every keyword it monitors. Ours did not, because we had built quality checks a week earlier — but only after burning eleven paid API calls on data we could not use, and only after catching ourselves about to alert a user on a drop that never happened.
The incident ended the way the reporting said it would: on day six the payloads came back clean, and a keyword that had been "absent" all week read position 1 on the very next check. Nothing about the site had changed. The pipeline had.
The Trap Nobody Warns You About: Partially Readable Results
You might assume a half-broken payload is better than a fully broken one. It is worse. Suppose the top two results in a poisoned payload happen to be real. Your site is at position 1, the tool reads it correctly, all good. Two days later a clean payload comes back and your site reads position 5. Your tracker now confidently reports a 4-position drop that is pure fiction — manufactured entirely by comparing one trustworthy reading against one untrustworthy one.
Partial visibility does not degrade accuracy gracefully. It converts it into plausible-looking wrong numbers, which are far more damaging than obvious gaps because nothing prompts you to question them.
How to Tell a Tracker Problem From a Ranking Problem
When a number looks wrong, run through these in order:
- Check the check's own timestamp. If the last successful check is older than your schedule, collection has stalled and everything since then is a display artifact, not a measurement. This is the single most overlooked signal.
- Compare against Search Console impressions, not GSC position. Impressions come from Google's own logs, on a completely different pipeline than any tracking tool. If impressions are flat while your tracker reports a collapse, distrust the tracker. (Compare impressions, not average position — averages blend personalization and SERP features and will "disagree" with any tracker for innocent reasons.)
- Look at the shape of the drop. Real ranking losses usually fade in over a few checks or track with an algorithm update. A clean keyword going from position 3 to "absent" in a single check — with no change to the page — is far more likely a pipeline event.
- Re-check manually, but know the limits. An incognito search is an independent reading, but it is also a different location, device and personalization profile, so it can legitimately differ by a few positions. It confirms or refutes gross absences ("not in the top 10 at all"), not small ones.
- Ask whether the tool can even see the depth it claims. Some plans and some API responses return fewer results than the tool's UI implies. If your tracker only reliably reads the top 5, its "outside top N" statement has a much wider error bar than its tone suggests.
What a Responsible Tracker Should Do — and What We Now Do
The incident rewrote our own bar for what "working" means. A tracker earns trust when it can state which of these three things is true at any moment: checked, nothing changed; checked, but the data was unusable, so we recorded nothing; and couldn't check at all. Those are different states with different meanings, and most tools collapse all three into "no change", which is exactly how fabricated history gets written.
The concrete rules we now run, which map one-to-one onto the checklist above:
- If a result payload contains links that cannot be resolved, the entire check is discarded — never written, never counted as an absence, never allowed near an alert.
- An absence only counts after two consecutive misses on the same keyword. One miss is a suspicion, not an event.
- Position depth is computed only from results actually visible in the payload, never from what the plan nominally covers.
- If unusable samples persist for days, the tool says so out loud — including telling you it is effectively flying blind, rather than charting silence as if it were data.
We wrote this from the tracking side because that is the side we live on. The same checklist works in reverse if you are evaluating a tracker: ask the vendor what happens when their data source returns garbage. "It doesn't happen" is the wrong answer. "We discard it and tell you" is the right one.
If you want a second pair of eyes on a few keywords that behave like this — quiet, alert-only, with unusable data discarded rather than charted — SerpCraft's free plan is built around exactly the rules above. It pairs well with Search Console as the independent cross-check, which is a combination we described more fully in our guide to tracking rankings affordably, and it sidesteps the disagreement problems covered in why rank tracking numbers never match.