← All posts

32 Search Console Fixes: 11 Worked, 5 Made It Worse

We marked 32 fixes on a live SaaS and measured every one against the 28 days before it. 11 improved, 7 went flat, 5 declined, and 9 can't be read at all.

QueryScope team · · 10 min read ·
search console gsc dogfood case study

Six weeks ago we published AppScreenshotStudio's Search Console numbers as a baseline: 277 clicks on 53,510 impressions, a 0.52 percent click rate, average position 10. Since then we marked 32 fixes on that site and measured each one against the 28 days before it. The same 28-day window now reads 549 clicks on 52,134 impressions, a 1.05 percent click rate, and a worse average position of 10.9.

Clicks nearly doubled on flat impressions while the position number went backwards. Below is the row-by-row scoreboard behind that: the 11 that improved, the 5 that declined, and the 9 whose verdict can't be read at all.

TL;DR:

  • Of 32 marked fixes, 23 have a readable verdict: 11 improved, 7 went flat, 5 declined. Only 3 are past their own settle window, so most of this scoreboard is still provisional.
  • The homepage retitle earned 81 percent more clicks while its click rate and its average position both got worse. Both are averages, and a wider query mix drags them down during a win.
  • Four of the largest click gains came with fewer impressions, not more.
  • Nine fixes can't be judged at all: three because the page is redirected or out of the index, four because there was no traffic before, two because they're inside the 14-day window.

What happened to 32 fixes in six weeks?

Twenty-three of the 32 have a readable verdict. Eleven improved, seven went flat, five declined. The other nine can't be read yet. And only three of the 32 are past their own settle window, which means most of what follows is a preliminary reading that will still move.

Every fix carries a recorded hypothesis: what changed, why, and the number we expected. That's what makes the misses legible instead of embarrassing. A trimmed view of four rows:

/                        improved   21 → 38 clicks
/tools/aso-audit         improved   13 → 25 clicks
/templates/active-bold   declined    2 → 0 clicks
/iphone-17-pro-mockups   no data    page not indexed

Two caveats belong on the whole table before any of it means anything. Most rows sit under 500 impressions, where a single click swings the rate by whole percentage points, so the direction is a hint and not a result. And the three settled rows are a small sample by any standard: two improved, one flat.

Why did clicks double while the average position got worse?

Because click rate and average position are both ratios, and a page that starts appearing for more searches drags both down while earning more. Google defines position as "the topmost position occupied by a link to your property or page in search results, averaged across all queries in which your property appeared." [1] Add queries at the bottom of page one and the average falls.

The homepage is the clean example. We retitled it to lead with the category term instead of the brand, because the head term was sitting on a weak subpage. Six weeks later:

  • Clicks: 21 to 38, up 81 percent.
  • Impressions: 174 to 348, roughly doubled.
  • Click rate: 12.07 percent to 10.92 percent, down 1.15 points.
  • Average position: 4.8 to 7.6, down 2.8 places.

Every number a dashboard paints green went the wrong way except the one that pays. The page didn't get worse at anything. It started showing up for a wider spread of searches, most of them further down the page, which is exactly what a category-led title is supposed to do. If you judge that edit on click rate or position, you revert it. The same arithmetic sits behind every metric Search Console reports, and it's why a fix gets judged on clicks when the query mix is moving.

Which fixes actually moved anything?

Eleven improved, and the ones that moved most were content edits rather than title rewrites. The pattern worth naming is that the direction was right more often than the reasoning was. In two of the biggest gains, the number we predicted and the mechanism we predicted were both wrong, and the page gained anyway.

  • The design guide gained +17 clicks against an expected +4. It went from 5 to 22 clicks, click rate 0.34 to 0.94 percent, position 8.5 to 7.3, on impressions rising from 1,463 to 2,350. The change was one added section answering a 2026 trends question an AI Overview was fielding above us.
  • The audit tool gained +12 clicks without reaching page one. The recorded expectation was "about 9 clicks a month if it reaches page 1." It went from 13 to 25 clicks while its position moved 17.0 to 15.8, so it never got near page one. The click gain was real and the stated reason for it was not.
  • Four of the gains came on fewer impressions. The build-free update guide went 27 to 39 clicks as impressions fell from 6,129 to 3,997. The automation post went 10 to 18 on impressions falling 2,931 to 1,534. The featuring guide went 10 to 20 on 3,965 falling to 2,655, and that one is settled.

Predicted click gains are an upper bound, not a forecast, for the same reason a striking-distance payoff is an upper bound: the estimate assumes a clean results page and can't see what sits above you. The useful part isn't the estimate. It's the gap between the estimate and what landed, which only exists because the number was written down before the edit shipped. That's the whole job of a click-rate queue that flags pages ranking well and earning nothing: not to promise the recovery, but to make the promise checkable.

Try it

Read your own Search Console, not just an essay about it.

QueryScope brings your real Search Console data into Claude Code or Cursor, so your agent reads it for you. From $14.99/month.

What do the five declines have in common?

Four of the five are too small to read. They sit under 500 impressions, where a single click moves the rate by percentage points and the verdict is noise wearing a label. We publish them anyway, because dropping them is how a scoreboard turns into a highlight reel.

  • A template page went 2 clicks to 0, impressions 248 to 120, position 7.7 to 9.9.
  • Another template page went 1 to 0, impressions 57 to 24, position 5.1 to 13.3.
  • A comparison page went 1 to 0 while its position improved, 7.8 to 7.0 on impressions falling 84 to 51.
  • The launch checklist went 5 clicks to 3 while its position improved, 8.9 to 7.6, on impressions falling from 2,758 to 1,712. This is the only decline with enough traffic to take seriously.
  • A de-indexed post went 3 to 1. The fix was a reindexing request, so the edit was "ask Google to look at this again" and the answer, so far, is no.

Two of those five ranked better and earned less, which is the mirror image of the homepage row. None of them is evidence that the edit caused the drop. Under 500 impressions there's nothing to read in either direction, and saying so is more useful than assigning blame to a change that happened to be nearby.

Why can't nine of the fixes be read at all?

Because a before-and-after needs a before, an after, and a page Google is willing to show. Nine rows are missing at least one. Three have no on-page data because the URL is redirected, consolidated, or not indexed. Four have no traffic from before the change, so there's nothing to compare against. Two are inside the 14-day window.

The most instructive row is the first kind. A device mockup page was marked fixed on 2026-07-18, and 23 days later it's still out of the index, so its experiment has produced no reading and can't produce one. It sits at the top of the site's queue with the reason attached: while the page is outside the index, its open fix cannot show any lift.

That's the index gate in a single row. Nothing you do to a page matters while Google declines to keep it, which is why finding the pages that never made it in comes before any of the other work, and why "crawled, currently not indexed" is its own diagnosis rather than a content problem. Seven fixes on this site were indexing fixes. One structural change, canonicalising four device variants onto each design, was marked on seven pages and produced five different outcomes: one improved, one flat, two declined, two with no data, one still waiting. Same decision, same day, seven verdicts.

Can you trust a before-and-after at all?

Only as correlation, and Google says so in its own words: "Keep in mind that there's no guarantee that changes you make to your website will result in noticeable impact in search results, and that positions in Google Search results aren't static or fixed in place." [2] Your before-and-after window contains every ranking movement you didn't cause, and there's no way to subtract them.

Three things follow, and we apply all three to the table above.

Windows contaminate. The tool's own headline for this site reads impressions down 17.3 percent against the prior 28 days. Read the monthly history instead and daily impressions are roughly flat, about 2,009 a day in July against 1,950 in August. The 17.3 percent is a window artifact, because the earlier window contained the peak. That's why the comparison at the top of this post runs against a figure published six weeks ago rather than a rolling one.

Settle time is per change type. A title and meta rewrite gets about 28 days. A content or link change gets about a quarter. Google's own advice on reading Search Console after an update is "waiting at least a full week after a core update completes before analyzing your site in Search Console," alongside a warning to "avoid doing 'quick fix' changes." [2]

The results page moved too. On 11 of the 17 flagged pages we could check, an AI Overview sits above the site, and four of those already cite it. Ahrefs measured that "the presence of an AI Overview now correlates with a 58% lower average clickthrough rate for the top-ranking page" across 300,000 keywords. [3] A flat click rate at a good position on this site is as likely to be the results page as the title, and a rewrite can't win back a click Google is keeping.

What the scoreboard changed about how we work

Five things, in the order they save the most time. Write the expected number down before the edit ships, because the gap between predicted and measured is the only part that teaches. Judge on clicks while the query mix is moving. Ignore anything under 500 impressions. Check the index before anything else, since a fix on a page Google won't keep has no verdict at all. And re-read a miss as a cue to reopen the hypothesis rather than as a failed edit.

Two honest limits, both about our own tooling rather than Google's. The site's flag history shows 16 click-rate flags of which zero became fixes and all 16 were accepted or let go, which is a fact about this site's results pages, not a verdict on the flag. And 16 earlier fix markers here carry no flag tag at all, because they predate the field, so the tool's own fixed-versus-accepted split under-reports real work. It prints that warning on the page rather than quietly reporting the flattering number.

None of this proves revenue. It measures clicks from Google and stops there, exactly as the baseline report six weeks ago said it would. If you want the loop itself, marking a change, waiting out its settle window, and reading the before-and-after without leaving your editor, that's what running Search Console in the terminal is for.

Sources

  • Site data: AppScreenshotStudio's own Search Console, read through QueryScope on 2026-08-10, covering the 28 days through 2026-08-09, with per-fix before-and-after windows anchored to each recorded fix date. The 2026-06-30 baseline is the previously published 28-day window through 2026-06-29.
  • [1] Google Search Console Help, What are impressions, position, and clicks? (position is the topmost result, averaged across all queries; impression counting).
  • [2] Google Search Central, Google Search's core updates and your website (no guarantee of noticeable impact, positions not static, wait a full week after a core update, avoid quick fixes).
  • [3] Ahrefs, Update: AI Overviews Reduce Clicks by 58% (300,000 keywords, December 2023 versus December 2025).
Try it

Read your Search Console where you code.

Ask your coding agent how your site is doing. QueryScope reads your real Search Console data in the terminal: clicks, queries, intent, and indexing. From $14.99/month. One to ten sites.