How Long Does a New Website Take to Rank? 26 Days Logged

On 2 August 2026 we shipped nine service pages on our own site and inspected them two hours later. Google had not seen a single one of them: 0 of 9 indexed. Eight days later, 7 of 9 were in the index. This piece is the rest of that month, from our first technical audit on 16 July 2026 to 11 August 2026, and it includes the parts that went wrong.

How long does it take for a new website to rank? Nearly every answer you can find quotes someone else's average. Averages hide the ordering of events, which is the part you need when you are staring at a site nobody visits. So here is the dated record from one real site, ours, with every number traceable to a file and the failures named rather than folded into a lessons list.

How long does it take for a new website to rank?

Longer than most pages will tell you, and our own 26 days produced one ranked US keyword and zero positions in the top ten. The published research points the same way: reaching the top of Google is rare, most pages never get there at all, and the results you are trying to displace are old.

Ahrefs' 2025 study of how long it takes to rank in Google found that 1.74% of newly published pages reach the top 10 within a year, rising to 6.11% once the sample is filtered to non-empty English content. Of the few pages that do get there, 40.82% arrive inside the first month. And 72.9% of the pages currently sitting in the top 10 are more than three years old.

Our sample is one site, so we are not competing with Ahrefs on scale. What we have instead is a log you can see: dated snapshots, the failures named, and the order things happened in. You can check the sequence rather than take an average on trust.

What did we measure, and what is this not?

One site, 26 days, no control group. Everything below comes from Google Search Console and a scripted tracker that writes dated snapshot files. The window runs from our first technical audit on 16 July 2026 to 11 August 2026. Two snapshots carry most of the numbers: 2 August and 10 August.

Read the click and impression figures carefully. Both are 28-day rolling windows, measured on 2 August and again on 10 August, eight days apart. They are not month one against month two. Twenty of the days inside those two windows are the same days.

One thing does work in this dataset's favour. Google has confirmed that a logging error stopped Search Console reporting impressions accurately between 13 May 2025 and 27 April 2026, and that the error touched impressions, click-through rate and average position while leaving clicks alone (Google, data anomalies in Search Console). Our window starts after the correction. Most new-site case studies published in the last year do not.

Treat this as a record rather than a study. It says what happened on one site and in what order, which makes it useful to compare against your own log and useless as a benchmark.

How long did Google take to index nine new pages?

Our nine pages were two hours old the first time we called them uncrawled, so the first delay in this story is ours, not Google's. The pages went live on 2 August 2026 at 13:01 and we ran URL Inspection two hours later. The verdict was 0 of 9 indexed. Eight days on, 7 of 9 were in.

At 18:02 the same day we deployed a fix that gave every one of the 46 URLs in our sitemap a real lastmod date, up from 11 of 46. By the 10 August snapshot, seven of the nine new service pages were indexed. Two were not: /grow/ai-sales-coach returned "URL is unknown to Google", and our AI SEO service page returned "Discovered, currently not indexed".

Nine new service pages, indexed URLs eight days apart

2 August 2026, two hours after launch

0 of 9

10 August 2026

7 of 9

We cannot tell you the fix caused the indexing. The lastmod change and eight days of page age arrived together, and one site with no control group cannot separate them. Our own log records it that way, and so does this article.

Google sets the expectation plainly in its own documentation on asking Google to recrawl your URLs: crawling can take anywhere from a few days to a few weeks, and requesting a crawl is no promise of inclusion in search results.

Discovery is only half of why pages sit outside the index. Indexing Insight monitored 1.7 million pages across 18 sites through the URL Inspection API and found that quality issues, not crawling, make up 88% of the indexing problems it recorded. Eighteen sites is a narrow base to generalise from, and worth knowing before you spend a week on your sitemap.

Why did impressions nearly double while clicks did not?

Because an average position around 30 collects impressions and almost no clicks. On the 28-day rolling window, impressions went from 1,688 on 2 August to 3,297 on 10 August, while clicks went from 12 to 19. That is neither a breakthrough nor a failure. It is what page three looks like from the inside.

Impressions, 28-day rolling window

Measured 2 August 2026

1688

Measured 10 August 2026

3297

Clicks, the same two 28-day windows

Measured 2 August 2026

12

Measured 10 August 2026

19

Click-through rate fell from 0.71% to 0.58% across those two snapshots. That is what happens when a small click count gets divided by a much bigger impression count. It tells you where the pages sit, not that something got worse.

The spread underneath did widen. Pages earning impressions went from 21 to 32. Our flagship guide to AI marketing agents went from 254 impressions to 457, while its average position for "ai marketing agent" moved from 31.8 to 30.9. A move of under a point at position 31 sits inside the noise in this data, so we are not counting it as progress.

That gap has a name in the wider data. Ahrefs looked at its own Search Console and watched the correlation between clicks and impressions fall from plus 0.425 in late 2024 to minus 0.352 in early 2025, which it called the great decoupling. Our sample is far too small to say anything about that trend.

We pointed two of our own pages at one search term

We built a page that competed with our own, and the search results had already told us it would. A check on 16 July 2026 for "ai development company" and "ai development services" in the US returned 5 of 7 identical URLs in the top ten. Every competitor ranking for both phrases covered them with a single page. N-iX ranked second for "company" with a page titled "AI development services".

Two of our pages, /build and its child on AI development services, were pointed at that single intent. The fix was to retire the child and 308 redirect it into the parent page. The commit shipped on 17 July 2026, the day after the check, and the next audit confirmed the redirect was live.

One caveat on that evidence. The raw search results were never archived, because this predates our vault. What survives is the overlap table in the decision note, so treat the 5 of 7 as our record rather than something you can re-run.

The cost was a page's worth of work. It also produced a rule we now apply before a page gets written rather than after it: check whether the search results already treat two phrases as one intent.

We put a number in a chart that our source did not contain

A draft of our own guide claimed 400 million ChatGPT users. The article we cited never contained that figure. Our editor caught it twice, once in the prose and once inside the chart, before the piece published on 3 August 2026.

What the source actually reports is 800 million weekly users in October 2025 and 900 million in February 2026. Those two figures are what the corrected chart in our LLM SEO guide carries now.

We are publishing this because a chart is the easiest place in an article to assert something no source supports. Readers argue with a sentence, and almost nobody re-checks a bar against the source it names.

Google mis-reported a number for about fifty weeks and disclosed it on its own data anomalies page. We mis-reported one for a single draft and disclosed that too. Disclosure does not undo either error, but it is the only reason anyone outside the room can check them, and the same goes for every number in this article.

The number that did not move at all

Editorial referring domains: zero on 2 August, zero on 10 August. Of every measure with a 90-day target attached, that is the one still sitting on the floor, and it is the reason this article exists.

Total referring domains did move, from 10 to 13. All ten at the 2 August count were URL shorteners or scraper sites, nine of them nofollow, spam score 50. We have no breakdown of the three that arrived after that. Nobody means domains like those when they say links. LLM brand mentions stayed at zero across both dates, and ranked US keywords stayed at one.

Day 9 of 90 reads like this: 19 clicks against a target of 100, one ranked keyword against 25, zero editorial referring domains against 10, and zero LLM brand mentions against one. The four targets are different sizes, so the chart below shows each one as a share of its own target: 19%, 4%, zero and zero.

Day 9 of 90, how much of each target is reached

Clicks, 19 of a target of 100

19%

Ranked US keywords, 1 of 25

4%

Editorial referring domains, 0 of 10

0%

LLM brand mentions, 0 of 1

0%

We would rather state the motive than pretend this arrived out of curiosity. A site with no editorial links needs something worth citing, and the only thing we had that nobody else was publishing was our own record with the failures left in. There is no ask attached to it.

What did we get wrong in our own record?

The file we created to stop age-blind conclusions stopped updating and missed three consecutive deploys. That file, the deploy log, exists only because an early audit described two-hour-old pages as never crawled. It stopped on 2 August 2026. We rebuilt it from git on 10 August and left the gap written into the file instead of closing it quietly.

Three more came out of the same re-audit on 10 August 2026. The drift baseline that is meant to be captured after every deploy had never once existed, because the folder it belongs in was never created. We captured the first one that same day. The monthly site-capture job failed on its first unattended run on 1 August 2026: it could not read its credentials file, changed nothing, told nobody, and was recorded as a success with exit code 0. Our vault also fails its own linter, with 10 problems in default mode and 21 in strict.

None of that changes the numbers above. It is here because a measurement record that never reports its own gaps is not much use to the person reading it.

What happens to this page in November?

Three things are true after 26 days, all of them measured. Nine pages went from 0 of 9 indexed to 7 of 9 in eight days, with the fix and the page age arriving together. Impressions on a rolling 28-day window went from 1,688 to 3,297 while clicks went from 12 to 19. And the number with a target on it, editorial referring domains, is still zero.

This is day 9 of a 90-day plan, so there is no 90-day result to give you. We will update this page with the 90-day record in November 2026, whatever it says. If a figure here looks wrong, the author is named on our leadership page, and any correction gets published here.

The full scorecard is below, including the rows that did not make it into the sections above. Both columns are single snapshots taken on the dates given, from Google Search Console and our own tracker.