Every Google SERP Change That Broke Search Data (2024–2026)

By Anurag Pathak · · 22 min read
An orange DETOUR road sign with a black arrow pointing left, standing beside a closed street

If you collect Google results for a living, the last two years have felt like driving a road where someone keeps moving the signs. A parameter you relied on for a decade stops working. A plain request that used to return results now returns an empty page with a 200 status. Then, in 2026, the links themselves stop pointing where they say they go.

None of these changes came with a migration guide. Most were spotted by rank-tracking companies before Google said anything, and a few Google has never commented on at all. That makes it hard to tell which change is behind the strange numbers in front of you.

This guide puts every significant change in one place, with dates and sources. For each one you get what changed, how to recognise it in the HTML or the URL, what it broke, and what it means for you. We compiled it from Google's own posts, court filings and the trade press, and checked every source on 12 November 2026.

Short answer: Since January 2025 Google has steadily raised the cost of collecting its results. It made JavaScript mandatory (January 2025), stopped honouring &num=100 (September 2025), began routing result links through opaque google.com/goto redirects (confirmed August 2026), and, according to rank-tracking firms, started serving swapped results to suspected bots (September 2026). Most of these fail silently, so check what you parse, not the status code.

The timeline at a glance

Here is every change in date order. "Status" is as of 12 November 2026. The sections below explain each one in plain language, then in technical detail.

DateWhat changedWhat it brokeStatus
1 Feb 2024"Cached" links removed from results; the cache: operator stopped working around 24 Sep 2024Archived copies of pages used for change trackingPermanent
14 May 2024AI Overviews launch to everyone in the US; a text-only "Web" filter (udm=14) appears the same monthThe assumption that position 1 is the first thing a searcher seesExpanded to 200+ countries
25 Jun 2024Continuous scroll removed on desktop, back to numbered pagesCollectors built around infinite scrollPermanent
Nov 2024"About X results" count moved under ToolsResult-count metrics read from the pagePermanent
15–17 Jan 2025JavaScript required to search; Google later named the system SearchGuardEvery plain-HTTP scraper (requests, curl, BeautifulSoup)Permanent
13 May 2025 – 27 Apr 2026Search Console logging error inflates impressionsYear-on-year impression comparisonsFixed going forward; old data not corrected
20 May 2025AI Mode launches to everyone in the US (udm=50)Coverage of a whole new answer surface200+ countries, 1B+ monthly users
11 Aug 2025Microsoft retires the Bing Search APIsThe main non-Google official search APIRetired
10–14 Sep 2025&num=100 stops being honoured; about 10 results per pageTop-100 rank tracking in one request; Search Console impression baselinesUnreliable (one unconfirmed sighting, Oct 2026)
19 Dec 2025Google sues SerpApi under the DMCA, naming SearchGuardLegal certainty for the SERP data industryAmended complaint dismissed 9 Nov 2026; Google has 21 days to refile
Jan 2026Custom Search JSON API closed to new customers; existing customers must move by 1 Jan 2027Google's only self-serve results APIClosing
19 May 2026New AI-first search box; follow-ups from AI Overviews continue into AI ModeClean separation between "results" and "AI answer"Rolling out
23 Jun –
26 Aug 2026		Result links routed through opaque google.com/goto?url= redirects; Google confirms 26 AugReading destination URLs from the pageLive, with fluctuations
From 13 Sep 2026Blocking wave, more CAPTCHAs and, per rank-tracking firms, swapped ("poisoned") resultsTrust in any single collected SERPOngoing, lower than its late-September peak
Late Oct 2026Limits on resolving many /goto links from one results pageFast, complete URL resolutionLive
Timeline chart from 2024 to 2026 marking cached links removed, AI Overviews, continuous scroll ending, the JavaScript requirement, AI Mode, num=100 ignored, the SerpApi lawsuit, the AI-first search box, /goto links and the September 2026 blocking wave
The changes cluster in 2025 and 2026, and they shift from removing features to hiding data.

Why Google keeps doing this

Google's public explanation is short and consistent. When it made JavaScript mandatory it told TechCrunch the change "allows us to better protect our services and users from bots and evolving forms of abuse and spam." When it confirmed the /goto links in August 2026, a spokesperson said: "We have a long history of deploying technical measures against evolving forms of abuse."

The wider context is AI. Search results are the raw material for AI answer engines and agents that need fresh facts, and Google's own AI Overviews and AI Mode compete with them. In its December 2025 post announcing the SerpApi lawsuit, Google said the scraping it was suing over "has increased dramatically over the past year."

Rank trackers and SEO platforms collect results in much the same way, so they get caught in the same net. Several of them have said so publicly. Keep that in mind as you read: very little of this was aimed at SEO tools, but SEO tools have felt nearly all of it.

2024: the quiet groundwork

Nothing in 2024 broke data collection outright. But five changes removed features that tools quietly depended on, and they set the direction for what followed.

Cached pages disappear (February and September 2024)

On 1 February 2024 Google's Search Liaison, Danny Sullivan, confirmed the "Cached" link was gone, calling it "one of our oldest features" that was meant for a time when "you often couldn't depend on a page loading" (9to5Google). The cache: search operator kept working for a while, then stopped completely around 24 September 2024, shortly after Google added links to the Internet Archive's Wayback Machine.

What it means: if you used Google's cache to see the version of a page Google had indexed, that view is gone. The Wayback Machine is now the usual substitute.

AI Overviews and the "Web" filter (May 2024)

Google rolled out AI Overviews to everyone in the US on 14 May 2024. The same month it launched a "Web" filter that shows only text links. Reporters quickly found it maps to the URL parameter udm=14, which Google has never formally documented. More on the AI layer below.

Continuous scroll ends (June 2024)

Google had added continuous scroll to mobile in October 2021 and to desktop in December 2022, showing "up to six pages of results" as you scrolled. It began removing it on desktop on 25 June 2024, with mobile to follow (MacRumors). Pagination was back.

The result count goes into hiding (November 2024)

The familiar "About 1,230,000 results" line moved under the Tools menu. Sullivan explained that "counts tend to be something researchers are more interested in," so moving them was a way "to keep them available but not be distracting" (Search Engine Roundtable). If you scraped that number as a competition metric, it is no longer on the default page, and it was always a rough estimate anyway.

January 2025: JavaScript becomes mandatory

In plain terms: Google stopped answering visitors that cannot run JavaScript. That one change ended the era of scraping Google with a simple HTTP request and an HTML parser.

When: SerpApi dates it to 15 January 2025, and TechCrunch reported it on 17 January. Google told TechCrunch that "fewer than .1%" of searches come from people with JavaScript turned off, so ordinary users barely noticed.

How to recognise it: you get an HTTP 200, not an error. The body is a shell page with a meta refresh to /httpservice/retry/enablejs, a "Please click here if you are not redirected within a few seconds" notice, and no result headings. We sent one plain request on 12 November 2026 and got exactly that: about 93 KB of HTML and zero <h3> result titles. Our earlier write-up, Google Search now requires JavaScript, shows the same result from July 2026.

$ curl -s -A "Mozilla/5.0" \
    "https://www.google.com/search?q=running+shoes" -o page.html
$ grep -c "<h3" page.html
0
$ grep -o '<meta[^>]*refresh[^>]*>' page.html
<meta content="0;url=/httpservice/retry/enablejs?sei=…" http-equiv="refresh">
# Status 200, about 93 KB, zero result titles: the JavaScript wall (12 Nov 2026).

What it broke: every tutorial built on requests plus BeautifulSoup, and every lightweight collector that did not render pages. On 17 January 2025 Search Engine Roundtable collected reports from tool vendors. Similarweb said the change "temporarily disrupted rank tracking and SERP data solutions across the industry, including ours." SE Ranking said some SERP features were "still unavailable for tracking."

What it means for you: collecting Google results now needs a full browser environment, which is heavier, slower and more expensive per page than an HTTP request. The worst part is the silent failure. If you only alert on HTTP errors, an empty 200 page sails straight through, a trap we cover in why your SERP scraper breaks at 3 a.m.

SearchGuard: the name behind the wall

In plain terms: for almost a year, nobody outside Google had a name for the system behind the JavaScript requirement. Google supplied one in its December 2025 lawsuit: SearchGuard.

What Google says it is: the complaint filed on 19 December 2025 describes it in a few short paragraphs:

What it means for you: the JavaScript wall, CAPTCHAs and later measures are best understood as layers of one system that Google keeps tuning. When one layer changes, collection quality can shift overnight without any visible change to the page a human sees.

CAPTCHAs and the /sorry page: the one loud failure

In plain terms: when Google decides traffic looks automated, it stops showing results and asks for a CAPTCHA instead. This is the oldest defence on the list and the only one that fails loudly.

How to recognise it: the request is redirected to a URL under https://www.google.com/sorry/index?continue=…, commonly with HTTP status 429 (Too Many Requests). The page says "Our systems have detected unusual traffic from your computer network." Google's help page for that message says it usually shows a reCAPTCHA and names search scrapers among the kinds of automated traffic that trigger it.

What changed recently: volume. In October 2026 Derek Perkins of the rank tracker Nozzle described "a significant increase in the use of captchas" as one of three tactics Google was using against tracking tools (Search Engine Roundtable).

What it means for you: a 429 is at least honest. You know you got nothing and can schedule a retry later. Our guide to Google's 429 "unusual traffic" error covers what that page means and the usual responses to it.

In plain terms: signed-out visitors from Europe can be shown a cookie-choice page before they see any results. A collector that does not handle it receives a page with buttons and no listings.

When: Google added an equal-weight "Reject all" button for signed-out users on Search and YouTube from 21 April 2022, starting in France and then across the EEA, the UK and Switzerland (Google). In March 2024 its Digital Markets Act changes added more than 20 Search changes in the EEA and new consent prompts for linking services.

How to recognise it: the final URL is on consent.google.com, or the page holds a form that posts back there. Google's cookie policy says it stores your choice in a cookie called SOCS, which "lasts for 13 months."

What it means for you: if you collect results for European markets, an empty result set may be a consent page, not an empty SERP. Results for EEA searches also differ in layout because of the DMA changes, so do not reuse US parsing assumptions for Europe.

September 2025: &num=100 stops working

In plain terms: for years, adding &num=100 to a Google search URL returned 100 results on one page. Rank trackers used it to check positions 1 to 100 in a single request. Google switched it off, so seeing the top 100 now takes about ten page loads instead of one.

When: Search Console impressions started falling on 10 September 2025, according to Brodie Clark's timeline. On 12 September Search Engine Roundtable reported the parameter working "about half the time." By 14 September it had stopped working for nearly everyone. A Google spokesperson's only comment was: "The use of this URL parameter is not something that we formally support."

How to recognise it: you ask for 100 results and get about 10. The start parameter (start=10, start=20 and so on) still pages through results.

What it broke: two things, one obvious and one surprising.

There is a second distortion to know about. Google later said a separate logging error "prevented Search Console from accurately reporting impressions from May 13, 2025 until April 27, 2026," and the fix applies only going forward (Search Engine Roundtable, 4 May 2026). So impression data from mid-2025 to April 2026 carries two overlapping problems. Clicks were affected by neither.

Current status: on 30 October 2026 Brodie Clark saw the parameter working "in some capacity" after clicking through from a knowledge panel, but Barry Schwartz could not reproduce it (Search Engine Roundtable). Treat it as gone.

What it means for you: decide how deep you really need to track. For most commercial keywords, the top 20 or 30 is where traffic lives. If your reports still compare impressions across September 2025, annotate the break. Our deep dives on the &num=100 removal and on why rank trackers and Search Console disagree go further.

AI Overviews and AI Mode: the answer moved above the results

In plain terms: Google now writes its own answer at the top of many results pages (AI Overviews) and offers a separate chat-style search (AI Mode). A "position 1" organic result can sit below a long AI answer, and AI Mode can answer a query without a classic results list at all.

The milestones

How to recognise them

AI Overviews appear inside the normal results page, above or among the organic listings, with linked sources. AI Mode is a separate surface. Clicking it produces a search URL carrying udm=50, a mapping documented by network-security vendors and SERP data providers rather than by Google. Several data providers also document that some AI Overviews finish loading after the rest of the page, so a capture taken too early can miss one that a person would see.

One reassurance from Google at I/O 2026: on the normal results page, "You'll continue to get a range of results… just like you do today." The classic page is not gone. It is just no longer the only page.

What it broke, and what it means for you

AI Overviews did not break collection the way the JavaScript wall did. They broke the meaning of rank. If you report positions without recording whether an AI Overview appeared, and which sources it cited, you are reporting half the page.

Search Console will not separate it for you either. Google's documentation says AI features, including AI Mode, are counted within the overall "Web" search type. Google added a generative-AI performance report (impressions only) on 3 June 2026 and extended it to all sites on 31 August 2026.

For the measurement side, see extracting AI Overviews at scale and our study of how AI Overviews and AI Mode cite different sources.

The official APIs close

In plain terms: while scraping got harder, the official doors also closed. That pushed more teams toward collecting results, which helps explain why Google's defences escalated.

If you still run on Custom Search, start planning now. Our Custom Search shutdown migration guide compares the options.

December 2025 onward: the SerpApi lawsuit

In plain terms: Google sued a company that sells Google results through an API, arguing that getting past SearchGuard breaks US copyright law. So far, the court has not agreed.

What it means for you: the court has twice found Google's DMCA theory insufficient, and the core holding about ordinary results is not coming back in this case. But a court case is not Google's only tool. The technical measures in the next two sections arrived while the case was running. This is not legal advice; our neutral explainer on the lawsuit has more context.

2026: /goto links replace direct URLs

In plain terms: Google started wrapping result links in its own redirect. The title and snippet still say "example.com," but the link points to a Google address with a scrambled code. Only by following the link do you learn where it really goes.

When

How to recognise it

Before the change, the href on an organic result was usually the destination itself. After it, the result looks like this:

<!-- Before (simplified) -->
<a href="https://www.example.com/running-shoes/"><h3>Best Running Shoes</h3></a>

<!-- After (simplified) -->
<a href="https://www.google.com/goto?url=CAES…"><h3>Best Running Shoes</h3></a>

The url value is not an encoded copy of the destination. ScrapingBee's 7 September 2026 analysis found the token Base64URL-decodes into a protobuf-like envelope whose payload appears encrypted, so "you cannot practically decode it back into the destination URL locally." Following the link with a normal GET returns a 302 whose Location header holds the destination. A HEAD request returned 200 with no Location header in its tests.

It is not limited to blue links. An SEO agency, Growtika, checked 493 links across eight US searches, covering organic results, videos, AI Overview sources, People Also Ask and ads, and found every one hid its destination.

What it broke

Brave's search team summed up the industry mood on 11 November 2026: "Scraping Google has become harder after the GOTO release, especially if results need to be fetched in real-time" (Brave).

What did not break: site owners' own analytics, as far as one test shows. In Growtika's test the visitor still arrived with a google.com referrer and Google Analytics still recorded "google / organic." Search Console reporting is unchanged.

What it means for you: check your stored URLs. If any contain google.com/goto, your domain matching and any downstream page fetches are quietly failing. And if your provider returns titles and positions but cannot give you destinations, your rank tracking is only as good as your title matching.

September 2026: blocking waves and swapped results

In plain terms: starting in mid-September 2026, rank-tracking firms reported that Google was not just blocking suspected bots but sometimes feeding them different results. The page looks complete and normal. The sites that usually rank are simply not there.

When

The numbers

Nozzle's study, summarised by PPC Land on 19 October 2026, covered 27.5 million SERPs from 49 countries collected between 13 September and 9 October:

Other vendors measured different rates, which is itself a clue that the treatment depends on how each collector's traffic looks. Ahrefs said it saw "between 5% and 20% of search results affected." SE Ranking said the issue "still affects roughly 2% of tracked results." Semrush said the impact on its users' rankings "has been limited." Google has not commented on any of it.

How to recognise it

You cannot spot a swapped page by looking at it. As Nozzle put it: "Nothing on the page gives it away. Each keyword's own history does." The working test, used in different forms by Nozzle and AccuRanker, is to compare today's top 10 with the sites that usually rank for that keyword. If most of them vanish at once, and especially if they come back the next day, distrust the page rather than reporting a ranking crash.

What it means for you: one collection is no longer proof of anything. Before you tell a client their site dropped out of the top 10, confirm it on a second collection. Our explainer on why Google results differ every time covers the normal sources of variation you should rule out first.

The pattern: failures got quieter

Line the changes up and a pattern jumps out. In 2024 Google removed features, which you notice immediately. From 2025 it started hiding data instead, and hidden data does not raise errors.

Diagnostic chart pairing six symptoms with their likely Google change: a 429 to /sorry/index means a CAPTCHA block; an enablejs page means the JavaScript requirement; a consent.google.com page means the EU consent interstitial; about 10 results instead of 100 means num=100 is ignored; google.com/goto links mean opaque result links; a full page missing the usual sites means swapped results
Only the CAPTCHA block returns an error. The other five look like a successful request.
What you seeWhat it usually meansLoud or silent
HTTP 429 and a redirect to /sorry/indexA CAPTCHA blockLoud: an error status
A page that points to /httpservice/retry/enablejsThe JavaScript requirementSilent: HTTP 200
A consent.google.com pageThe EU consent interstitialSilent: HTTP 200
About 10 results when you asked for 100num=100 is ignoredSilent: HTTP 200
google.com/goto result linksOpaque result linksSilent: HTTP 200
A full page missing the usual sitesSwapped resultsSilent: HTTP 200

That table is the most useful thing in this guide, because it tells you what to assert on. An HTTP status tells you almost nothing about Google any more. What you parsed tells you everything.

Here is the same logic as code. The first function labels a fetched page by the change most likely affecting it. The other two run the history check for swapped results. We ran it on 12 November 2026 with Python 3.14 against labelled sample inputs, plus the real JavaScript-wall page from the curl test above, which it labels correctly. It makes no network requests, so plug it in wherever your pages arrive (Python 3.9 or later).

import re
from collections import Counter
from urllib.parse import urlparse


def classify_google_page(status: int, final_url: str, html: str) -> str:
    """Label a fetched Google results page by the change most likely affecting it."""
    url = final_url.lower()
    body = html.lower()

    if status == 429 or "/sorry/" in url:
        return "blocked: unusual-traffic / CAPTCHA page"
    if "consent.google." in url or 'action="https://consent.google.' in body:
        return "consent interstitial (no results)"
    if "enablejs" in url or "enablejs" in body or (
        "http-equiv=\"refresh\"" in body and "<h3" not in body
    ):
        return "JavaScript wall (no results)"

    hrefs = re.findall(r'<a[^>]+href="([^"]+)"', html)
    results = [h for h in hrefs if h.startswith(("http", "/goto", "/url"))]
    if not results:
        return "empty: zero result links"

    opaque = [h for h in results if "/goto?" in h]
    if len(opaque) / len(results) > 0.5:
        return f"opaque /goto links ({len(opaque)} of {len(results)})"
    return f"ok ({len(results)} links)"


def usual_domains(history: list[list[str]], min_share: float = 0.6) -> set[str]:
    """Domains that held a top-10 spot on at least `min_share` of past collections."""
    counts = Counter(d for page in history for d in set(page))
    return {d for d, n in counts.items() if n / len(history) >= min_share}


def looks_swapped(today: list[str], history: list[list[str]]) -> bool:
    """True when under half of the usual top-10 domains are present."""
    usual = usual_domains(history)
    if not usual:
        return False
    kept = len(usual & set(today)) / len(usual)
    return kept < 0.5


def domain(url: str) -> str:
    return urlparse(url).netloc.removeprefix("www.")

Run against five sample pages and two sample top-10 lists, it prints:

blocked: unusual-traffic / CAPTCHA page
consent interstitial (no results)
JavaScript wall (no results)
opaque /goto links (2 of 2)
ok (1 links)
False   # 4 of 5 usual domains still present
True    # usual domains replaced by reddit.com, facebook.com, quora.com ...

The 60% "usual site" share and the 50% threshold mirror the definition in Nozzle's published study. Tune both to your own data. The point is to make every silent failure loud before it reaches a report. Keep the raw page alongside each parsed result so you can check a suspicious one later.

What this means for you

The right response depends on what you do with the data. Here is the short version for the four groups this hits hardest.

If you run rank tracking or SEO reporting

If you build AI agents or RAG pipelines

If you do research or market analysis

If you mainly read Search Console

Where Serpent fits

If you would rather not maintain any of this yourself, that is what a SERP API is for. Serpent's Google SERP API returns Google results as structured JSON from one HTTPS call: organic results, Featured Snippets, People Also Ask, ads, related searches and optional AI Overviews with their sources. You choose from 112 country codes, and there is no browser, proxies or HTML parser for you to maintain.

Rank-tracking teams can build on the rank tracking API, and you can try any query in the playground before writing code. Pricing starts from $0.03 per 1,000 Web units, with no subscription; the documentation lists every field and parameter.

import requests

resp = requests.get(
    "https://apiserpent.com/api/search",
    headers={"X-API-Key": "sk_live_your_key"},
    params={"q": "best running shoes", "engine": "google", "country": "us"},
)
for r in resp.json()["results"]["organic"]:
    print(r["position"], r["title"], r["url"])

Google results as JSON, without the upkeep

One API call returns organic results, Featured Snippets, People Also Ask and optional AI Overviews in clean JSON. New accounts include free calls, then pay-as-you-go pricing with no subscription.

Get Your Free API Key

Explore: Google SERP API · Rank Tracking API · Pricing

FAQ

What are Google's /goto links?

Since mid-2026, Google has routed many search result links through a redirect of the form https://www.google.com/goto?url=CAES… instead of linking straight to the destination. The value is an opaque token, so the destination is not readable in the page. Google confirmed the rollout on 26 August 2026 as one of its technical measures against abuse.

Does &num=100 still work on Google?

Not reliably. Google stopped honouring it around 10 to 14 September 2025 and told reporters the parameter is not something it formally supports. On 30 October 2026 one SEO reported seeing it work in a narrow case, but others could not reproduce it, so plan on about 10 organic results per page.

Why did my Search Console impressions drop in September 2025?

Mostly because rank trackers stopped loading 100 results at a time, so deep listings stopped collecting automated impressions. Separately, Google says a logging error inflated impressions from 13 May 2025 to 27 April 2026 and that historical data will not be corrected. Clicks were not affected by either.

What is a poisoned or swapped SERP?

It is a complete, normal-looking Google results page in which most of the sites that usually rank for the keyword have been replaced, often by forums and social platforms. There is no error or CAPTCHA. Rank-tracking firms began reporting it after 13 September 2026; Google has not commented.

Is SearchGuard the same as a CAPTCHA?

No. SearchGuard is the name Google used in its December 2025 lawsuit against SerpApi for its anti-bot system, which it says launched in January 2025 and works partly by sending a JavaScript challenge to unrecognised traffic. The complaint describes CAPTCHAs as another aspect of SearchGuard.

Can I still get Google results without running my own scraper?

Yes. Google's Custom Search JSON API is closed to new customers and ends for existing ones on 1 January 2027, so many teams now use a commercial SERP API. You send one HTTPS request and receive structured JSON, with no browser, proxies or HTML parser of your own to maintain.

References & Further Reading

Related Posts

Google Killed &num=100: Rank Tracking Costs 10× More

Google Killed &num=100: Rank Tracking Costs 10× More

What changed on September 11, 2025, why Search Console impressions fell off a cliff, and the two ways to get the top 100 back.
Google Search Now Requires JavaScript: What Died in 2025

Google Search Now Requires JavaScript: What Died in 2025

Why a plain HTTP request to Google returns a 200 shell page with zero results, and the realistic options for pipelines that depended on it.
Google's Legal Action Against SerpApi: What It Means

Google's Legal Action Against SerpApi: What It Means

A calm, neutral walkthrough of the case, what the court has ruled, and what it does and does not change for your integration.
Why Your Rank Tracker and Search Console Never Match (2026)

Why Your Rank Tracker and Search Console Never Match (2026)

The measurement differences behind the gap between tracked positions and Search Console averages, and how to reconcile them.