**TL;DR** — Across 19 client sites in April 2026 we measured time-to-first-citation (TTFC) for newly published pages: how long between a page going live and its first appearance inside a Google AI Overviews answer card. Our cohort median is 11 days; the 90th-percentile page takes 41 days, and roughly 18% of pages never reach a citation inside the 90-day window we track. Three operational moves shorten TTFC measurably — IndexNow submission inside five minutes of publish, an internal link from a page that is already being cited, and tight publication-time consistency between the sitemap `lastmod`, the canonical URL and the page's own published timestamp. None of this is exotic. The gap between the median and the long tail is almost entirely operational, not editorial.
Why we started measuring TTFC
Citation tooling in 2026 is built around binary outcomes — is the page cited, yes or no — and weekly aggregates. That is sufficient when you are asking whether the AI search channel is alive at all. It is not sufficient when the question becomes "how do we beat a competitor to the answer slot on a query they will publish about next week." In a steady-state SERP, latency does not matter much — once everyone is cited, the order of arrival is forgotten. In AI Overviews, latency matters because the composer's first set of citations on a fresh query becomes the gravitational centre of the answer for weeks afterwards; arriving second is materially worse than arriving first.
There is a second motivation: most of the operational work that affects TTFC is invisible inside the editorial team. The page goes live; nobody can tell from the CMS dashboard whether it was submitted to IndexNow, whether the canonical link points at the right URL or whether the sitemap got regenerated. A latency metric makes the operational gap legible — when the median is 11 days and one team's pages consistently cite at 25, the question stops being "is the content good enough" and starts being "what is your publishing pipeline doing differently."
How we ran the measurement
Nineteen client sites — eight SaaS, six DTC, three publisher and two B2B services — across April 2026. Every new page got a publication timestamp captured to the second from the CMS event log. We ran a Playwright basket of 60 commercial-intent queries per client every day at 09:00 UTC and 21:00 UTC, capturing the AI Overview answer card and the cited URL set. TTFC equals the timestamp of the first capture in which the new page appeared as a cited URL minus the publish timestamp. We dropped any new page whose target query was not part of the standing basket — without a query covering it the metric is undefined, and a third of new pages fell into this hole.
Two normalisation moves matter. First, we used a 90-day measurement window per page; pages older than 90 days that had not yet been cited were treated as right-censored. Reporting the mean TTFC without that censoring would make the metric look better than it is — the long tail does most of the damage. Second, we excluded pages where the citation event happened in the same hour as publish; AI Overviews rarely indexes that fast, and on inspection these were always cases where the page was a republish of an already-cited URL with the same canonical, not a true first citation.
The shape of the TTFC distribution
The distribution is bimodal in a way the median number hides. About 41% of pages got their first citation inside the first 14 days; 23% reached it between days 14 and 41; the remaining 36% either trailed deep into the 90-day window or never cited at all. There is a cliff at around day 60 — pages that have not been cited by then almost never recover inside the window. Once a page crosses the citation threshold it tends to stay cited; once it has been ignored for two months it tends to stay ignored. The composer is not running a continuous lottery; it is running an early-evaluation cycle and an opt-out for everything that misses the cycle.
Pages that took longer than 41 days to cite were also more likely to lose the citation later — re-citation persistence on those pages was 64% over the next 30 days, vs. 89% for pages cited inside the first 14 days. Late citations are less load-bearing than early ones, which means the cost of slow TTFC is not only "later traffic" but also "less durable traffic." The metric to optimise is the early band, not the long-run average.
Lever one: IndexNow submission inside five minutes of publish
Pages submitted to IndexNow inside five minutes of publish had a median TTFC of 6 days. Pages discovered by the crawler without an IndexNow ping had a median TTFC of 19 days. The mechanism is straightforward — IndexNow tells Bing (and through Bing's relay, several other engines) that a URL exists; the AI search composers reuse that index, and the discovery latency drops from "next sitemap re-crawl" to "next composer pass." The work is small; the leverage is large; we still find roughly half of audited sites have IndexNow disabled or wired to a publisher account that is not connected to their actual publish pipeline.
A subtle gotcha: the IndexNow ping has to come from the publish event itself, not a nightly cron. We audited four clients whose IndexNow setup was technically correct but ran daily — the median delay between publish and ping was 11 hours, which negated almost all of the TTFC benefit. Move the ping into the same webhook that fires when an editor clicks Publish, and you get the full effect. Move it to the next sitemap regeneration cron, and you might as well not have it.
Lever two: an internal link from an already-cited page
Pages that had at least one internal link from another page that had been cited within the prior 30 days had a median TTFC of 7 days. Pages with internal links only from pages that had never been cited had a median TTFC of 18 days. The signal is again mechanical — the composer re-fetches cited pages much more often than uncited ones, and a fresh outbound link inside a frequently-fetched page accelerates discovery for the linked URL. The internal-link best practices that mattered for traditional SEO matter again here, but with a different ranking criterion: the value of the linking page is no longer "PageRank" but "recent citation frequency."
Practically, this changes how we structure pillar pages. We used to recommend "evergreen pillars that link to seasonal content"; we now recommend pillars that get edited every two to three weeks specifically to add and rotate links to the most recent publishes. The pillar earns a steady citation rate; the new content piggybacks on the pillar's fetch cadence; the TTFC of the new content drops by roughly half. Editing the pillar costs a few minutes and is the highest-leverage half-hour in any weekly editorial review.
Lever three: publication-time consistency
Pages where the sitemap `lastmod`, the canonical URL response header, the page's own JSON-LD `datePublished`, and the actual publish event all agreed within five minutes had a median TTFC of 9 days. Pages where any two of those signals disagreed by more than an hour had a median TTFC of 22 days. The composer treats publication-time disagreements as low-confidence signals; when the page says it was published an hour ago but the sitemap claims yesterday, the composer falls back on the older timestamp and the freshness boost is wasted.
The most common failure mode we see is a sitemap regeneration cron that runs every hour or every six hours. That introduces a built-in lag between actual publish and sitemap `lastmod`, which the composer reads as evidence the page is stale. The fix is event-driven sitemap regeneration: the publish webhook also rebuilds (or patches) the sitemap entry, so the lastmod matches the publish timestamp within seconds. This change alone moved one client's median TTFC from 16 days to 8 days inside one month, with no editorial change.
The 18% that never get cited
Three buckets explain almost all of the never-cited cohort. The largest is queries the composer never triggers an AI Overview on — about 40% of the never-cited pages target queries where AI Overviews have not yet been enabled in the relevant locale. Citation is impossible because there is no answer card. The fix is editorial: pick queries that already trigger AI Overviews in the basket. Pages targeting queries with stable AI Overview presence cited at a 4× higher rate than pages targeting queries where AI Overviews appeared on fewer than half of basket captures.
The second bucket is content that does not pass the composer's quality bar — thin pages, heavily templated programmatic pages with little unique content, and pages whose entire body is a paraphrase of a single competitor source. The composer's skip behaviour on these is binary; once it decides a page is low-quality it does not re-evaluate later. The third bucket is structural error: noindex tags accidentally left on by the dev team, X-Robots-Tag headers from a CDN rule, soft-403 walls on AI bots from a misconfigured WAF. We caught all three on different clients in April. Each was fixable in under an hour; each had been silently breaking TTFC for weeks.
What changed in our publishing checklist
Three operational additions. We require IndexNow ping inside the publish webhook, with the ping going out within five minutes of the page being live and verified by reading the IndexNow indexing report two hours later. We require every new commercial-intent page to receive at least one inbound internal link from a page cited within the prior 30 days, added at publish time rather than during a quarterly internal-link audit. And we require the publish webhook to also patch the sitemap and any cached canonical, with a deploy hook that fails the build if the lastmod and the page's `datePublished` disagree by more than five minutes.
We dropped one habit. Through 2024 and 2025 we treated weekly sitemap regeneration as best practice; in 2026 it is too slow. The latency cost of a six-hour or 24-hour sitemap-regeneration window is roughly half the median TTFC, and it is the single easiest piece of operational debt to repay. Rebuild on publish, not on cron.
- 01Measure time-to-first-citation per page, not just citation rate. Median 11 days, P90 41 days, ~18% never cite inside 90 days in our cohort. The long tail and the never-cited cohort are where the operational gaps live.
- 02Wire IndexNow into the publish webhook itself, not a cron. The TTFC delta between sub-five-minute pings and once-daily pings is roughly 13 days in our sample.
- 03Add at least one internal link from a recently-cited page on every new commercial-intent publish. Pages without one took median 18 days vs 7 days with one.
- 04Make the publish event also rebuild or patch the sitemap. Lastmod that disagrees with `datePublished` by more than an hour costs roughly two weeks of TTFC.
Where this argument breaks
For sites publishing fewer than four pages a month, the TTFC distribution is too sparse to read; the metric becomes noisy and the audit overhead is hard to justify. For programmatic sites with thousands of templated pages, the average TTFC is dominated by the long tail and the median hides more than it reveals — those teams need a per-template TTFC view, which is a different analysis. In Chinese-language search the discovery cadence is set by Baidu's own pipeline rather than IndexNow; the structural advice transfers but the tooling does not, and we run a separate TTFC audit for that cohort. Outside those carve-outs, TTFC is the metric most clients did not know they should be tracking — and it is also the metric whose gap from median to long tail is the easiest piece of operational ground to recover.