--- title: "How Rankly measures what a change did · Rankly" description: "Each change is read against similar pages left alone, from the day Rankly sees it live: at 28 days, 90 days and a year. The rules, and when a reading can’t tell." url: https://rankly.cc/measure/ --- # How Rankly measures what a change did Each change is read three times after it goes live. With 5 or more similar pages left alone, it is read against them, so a good or bad month for the whole site isn't counted as the change's doing. With fewer, it is read before and after only. Change #5, the pricing guide on theagentsindex.com: its five dates. 1. Handed over 7 Nov 2. Seen live 7 Nov The clock starts 3. 28 days 5 Dec First reading 4. 90 days 5 Feb 2027 Second reading 5. 1 year 7 Nov 2027 Last reading ## What it reads before it asks for a change. Google and Bing for every page and search, each live page, and AI answers. ### What Rankly read on theagentsindex.com this month The change each became - Times shown in Google /best-ai-agents, 28 days 457 A new page title#150 - Place in Google its average for those searches 5 A new page title#150 - Pages Google has never crawled of the 614 pages in its sitemap 197 Links to 20 of them from pages Google reads#125 - Pages Bing has never visited handed over, 7 more held back to compare 50 Handed to Bing#191 - AI answers citing the site answers read from 10 to 13 Nov 4 of 94 An order to open /eximagent by saying what it is#217 ### Where a click claim comes from /best-ai-agents was shown 457 times in 28 days at an average place of 5, and clicked once (0.2%). Rankly expects about 6.3% at those places, from public click studies by position, read on the low side. | Search | Shown | Place | Expected | | --- | --- | --- | --- | | ai coding agents 2026 | 104 | 3.8 | 7.8% | | ai coding agent 2026 | 79 | 4.2 | 6.6% | Searches by people only: rank-checking tools left out. ### One AI-answers reading From 10 Nov to 13 Nov, 27 questions about AI agents gave 94 answers. theagentsindex.com was cited in 4. 4 of 94 answers citing the site - ChatGPT 67 - Gemini 17 - Google AI Overviews 10 Answers by engine. ChatGPT is asked every question each week; Perplexity, Gemini and Claude take turns. ## A reading closes one of four ways. 1. Worked The changed page gained 10% or more on similar pages left alone, beyond their normal spread. Kept, and read again later. 2. No effect Within the normal spread of the pages left alone, or a real change smaller than 10%. Left as it is, and read again later. 3. Hurt The changed page lost 10% or more against the pages left alone, beyond their normal spread. Rankly hands over the old text to put back. 4. Can't tell yet Under 30 of what it measures (clicks, times shown or pages indexed) across both windows, or fewer than 5 pages left alone. Its before and after are shown, with no verdict. ### What each kind of change needs for a verdict Pages indexed, clicks and times shown A 10% move against the pages left alone, with at least 30 of what is measured across both windows and 5 pages left alone. Place in Google Scored as 100 minus the place, so place 1 is 99, then read the same way as clicks. Links At least 2 new sites linking to the changed pages, more than the pages left alone gained. Cited in an AI Overview At least one more search whose Overview cites the page, and 25 points more of them than for similar searches. Share of AI answers At least 10 answers in each window and a 10-point rise against the comparison questions. Fewer than 5 comparable pages Read before and after only, with no verdict. How a 28-day reading looks Google clicks a day on a changed page, against 24 similar pages left alone. A sample, not one of our changes 1. Worked +41% against +3% Sample a lift of 37% over pages left alone Seen live Kept. Read again at 90 days. 2. No effect +5% against +6% Sample within the normal spread of pages left alone Seen live Left as it is. Read again at 90 days. 3. Hurt \-32% against +2% Sample a fall of 33% against pages left alone Seen live Put back: Rankly hands over the old title and description. Drawn to the same rules as a real reading. Our first readings are due from 5 Dec. ## When it can't tell, it says so. Too little measured, too few pages to compare with, or a Google update in the window. 1. ### Too little to measure Due 10 Dec 1 of 30 clicks across both windows /best-ai-agents had 1 click in the 28 days before its new title. Under 30, its reading shows before and after, with no verdict. #150, theagentsindex.com 2. ### Too few pages left alone Due 11 Dec 3 of 5 similar pages needed to compare with Hand 31 pages to Bing that Bingbot had never fetched: only 3 pages like them were left alone, so it is read before and after only. #185, another site we run 3. ### A Google update Marked A reading is marked, not judged, when a Google update ran through a quarter or more of its window. An update moves every page at once. The measured change is kept, with the mark. ## Where our changes stand. None has been read yet. First readings 5 Dec. Readings taken while Google’s November update was running are marked, not judged; the first judged readings are due in February 2027. **234 changes** written out across the 13 sites Rankly runs on, since 7 Nov. - **161** applied and seen live - **66** ready, waiting for the site's team - **5** refused, each with its reason - **2** withdrawn by Rankly, with its reason