Zero to Cited: how a new site climbs into AI search

1. The result so far, and its limits

This page began as a plan and is maintained as a log. It is one site's attempt at the discipline defined in what generative engine optimisation is. What it can honestly report is a sequence: the order in which each named crawler arrived, and the order in which the first citations appeared. What it cannot report is a latency figure anyone should plan against, because one site produces one trajectory, and a trajectory is not a distribution.

That limitation is worth stating up front rather than in a footnote, because the genre this page sits in routinely does the opposite. "New sites get cited in X weeks" is a claim that requires many sites, measured the same way, with the failures included. This is one site, publishing its own numbers as they arrive, on the grading scheme described in the provenance audit and the measurement standard.

The strategy being run, in one line: get into Bing's index immediately, publish original data, seed brand mentions, and keep everything fresh — then let the fast engines cite you while authority slowly compounds for the slow ones. The reason that is viable at all for a new domain is a measured shift in where AI Overview citations come from.

That decline is independently reported and tracked in more depth on the AI Overview statistics hub. AI citation no longer requires top-10 Google rankings, so a new site can be cited long before it "ranks" in the traditional sense.

2. Which engine cites a new site first?

Not all AI engines are equally reachable. The differences come down to how each one sources content, and they produce a difficulty gradient a new site can sequence against.

The transferable finding is the ordering, not the timing: Perplexity and Bing first, AI Overviews later, ChatGPT last. That ordering follows from how each engine sources — on-demand crawling and a push index reward a new site quickly, while an authority gate does not. No latency figure should be read from it: this is one site, n=1, with no control group.

Sequence against that gradient: win Perplexity and Bing first, treat AI Overviews as a month-3-plus goal, and treat ChatGPT as the long game you are compounding toward rather than the thing you measure success by in month one. Claude, Gemini and the delta between Google's two AI surfaces each behave differently again — a spread visible in platform citation-pattern analyses and overlap measurement.

Perplexity is the easiest first citation. ChatGPT is the last domino to fall. Don't judge month two by ChatGPT.

3. The milestone timeline

The phase-by-phase climb below assumes a focused, well-executed niche site — not a mega-brand with existing authority. Read the months as a rhythm, not a promise. They describe a shape consistent with the published evidence on indexing speed and engine behaviour, and with what this site has recorded; they are not a measured latency estimate from any sample.

The phases overlap in practice — a first Perplexity citation often lands while you are still fighting to get fully indexed in Google. Anchor yourself to the leading indicators, indexation and crawler hits, rather than the lagging ones. How long the gap between them actually runs is what the crawl-to-citation latency study exists to measure across more than one site, and whether a citation survives once won is the half-life study. If you launch, publish three thin pages and wait, none of this happens: discovery and indexing still occur, but nothing gets cited because nothing is citable.

One caveat on how fast things can move in the other direction. Profound's case study of Ramp reports a move from 3.2% to 22.2% AI visibility in a single month — a 594% lift and 300-plus new citations — through focused optimisation. Ramp had existing authority, so that is not a new-site trajectory. It does show that citation share can step-change in a way Google rankings generally do not, because citation is gated more by extractability and freshness than by accumulated trust.

4. Fast levers vs slow levers

Knowing which is which keeps you from spending month one on things that cannot pay off until month twelve.

The fast levers, in detail:

• Bing indexing via IndexNow — the cheapest speed win, and it unlocks ChatGPT via Bing's index, which matters more than Bing's own market share would suggest. The protocol is open, widely adopted, and Microsoft has said plainly that sitemaps still matter for AI-powered search.
• Original data and stats-dense content — reported cited at 3–10× the rate of standard posts, and the one lever with controlled evidence behind it. Which tactics have any evidence at all is graded in the tactic scoreboard.
• Brand mentions — reported as the strongest correlate of AI Overview visibility in vendor data, at roughly 0.664 versus roughly 0.218 for backlinks. That is a correlation with an uncontrolled brand-size confound, not a demonstrated cause; seed mentions because they are cheap and defensible, not because the number proves anything.
• Freshness cadence — around half of AI-cited content is under 13 weeks old, under-30-day pages are reported alongside ~3.2× more citations, and 76% of ChatGPT's most-cited pages had been updated within 30 days. All observational.

The slow levers are real and patient. ChatGPT citations are gated by referring-domain authority. Domain authority itself accrues slowly and sets the ceiling. And Google top-10 organic rankings are slowest of all: only 1.74% of new pages reach the top 10 within a year, and the average #1 page is five years old. Do not gate an AI strategy on any of the three.

5. Why ChatGPT lags, and why that is fine

Analyses of how ChatGPT selects sources put domain authority at roughly 40% of the weight in browsing mode. They report that sites with 32,000-plus referring domains are about 3.5× more likely to be cited than sites with under 200. A brand-new domain is authority-starved by definition.

The response is not to chase ChatGPT directly. Win the engines that do not gate on authority — Perplexity, Bing, and AI Overviews via the 11–100 range — while links accumulate. Because ChatGPT reads from Bing's index, the Bing work is quietly building the eventual ChatGPT presence anyway. Author signals may matter here too, though the E-E-A-T study is an attempt to find out rather than an assertion that they do.

The practical consequence is a mindset one. A founder who checks ChatGPT in week two and sees nothing often concludes the strategy failed. ChatGPT is the last engine that will cite a new site; its silence says nothing about the Perplexity and Bing progress happening underneath.

6. The metrics to track weekly

You cannot see AI citations in Search Console, so you need a purpose-built dashboard.

Report the leading indicators weekly and the citation counts monthly. AI answers vary between identical runs, so a weekly citation number is mostly noise, while weekly crawl and indexation counts are stable enough to act on.

The tooling stack is smaller than it looks: Search Console and Bing Webmaster Tools, your raw server logs, and a spreadsheet of 15–30 target questions run through each engine on a fixed schedule. That manual panel is the ground truth every paid AI-visibility platform approximates. Buy one only when you have movement to measure. The agents themselves are documented by OpenAI and Perplexity, and Cloudflare's breakdown shows crawl volume at network scale.

7. Confounds in your own timeline

If you run this plan and get cited, you still will not know why. That is not defeatism; it is what a single-site, no-control-group observation can support, and it applies to this site's log exactly as much as to yours. Open question Several things move at once during a launch.

• You changed many things at once. A launch ships schema, a sitemap, IndexNow and new content in the same week. Attributing the result to any one is guesswork.
• The engines changed too. Sourcing behaviour shifts without notice. A citation appearing in month three may reflect an engine update, not your work.
• Your query set drifted. If you added prompts over time, later months look better for a purely mechanical reason.
• Seasonality and news. A topic that spikes in the press gets more retrieval attention for everyone in it.
• Sampling noise. AI answers are non-deterministic; the same prompt run twice can cite different sources.

The cheap fix is a control. Pick two comparable pages, optimise one, deliberately leave the other alone, and track both on the same panel. You still will not have a clean experiment, but you will have something better than a story.

8. How this page gets misread

The months are a promise. They are not. Your niche, competition and execution move every number, and n=1 cannot produce a benchmark.

The modelled charts are measurements. The two charts here are labelled as models, and that label is load-bearing. They show a relative ordering between engines, not counts anyone observed on a real new site.

Silence means failure. A new site is meant to be quiet in ChatGPT for months. Reading that as a verdict is the most common way people quit early.

A crawl is a citation. A bot hit means you were fetched, not quoted. Fact Keep the two metrics in separate columns, or you will report growth that has not happened.

If you take one habit from this section, take this one: before you report a number, say out loud what it would look like if the number were wrong. If you cannot answer, you are not measuring yet.

9. Null results we would publish

A plan that can only produce good news is not a plan. Stated before the data arrives:

• No first citation by month six on a fully indexed, well-structured, freshly updated site. That would contradict the shape this page describes.
• Citations arriving in the opposite engine order — ChatGPT first, Perplexity last. That would undermine the difficulty gradient in section two.
• Crawler cadence rising with no citation change at all over a long window, which would weaken the case for crawls as a leading indicator.
• Original-data pages performing no better than ordinary posts on the same site — the most interesting null of the set.
• A control page left untouched performing the same as an optimised one, which would suggest the optimisation was not the driver.

Open question Writing these down in advance stops success being redefined after the fact. If any happen here, they get published as plainly as a win would — in the null results registry, alongside every other method and dataset in the studies index.

10. Who this applies to, and who it does not

It fits a new, focused, content-led site in an informational niche with real query volume, not one dominated by entrenched giants. It fits much less well elsewhere. AI Overviews appear on roughly 16% of queries overall — Google documents the surface here — but that ranges from high on informational and comparison queries to near-zero on transactional and navigational ones, so a commerce-heavy, local or YMYL niche has a different surface to win. A regulated field may find engines reluctant to cite anyone but recognised institutions. And an established domain skips the discovery phases entirely, so these month markers would understate it badly.

Hypothesis The underlying claim is that engine ordering is more stable than engine timing. The sequence — fast engines first, authority-gated engines last — should hold across most niches. The calendar attached to it should not be treated as portable, and this site's own n=1 log is not evidence that it is.

11. Your day-0 plan

1. Ship with a clean sitemap.xml and a robots.txt that allows the AI search bots — the blocking census shows how often sites get this wrong by accident. Skip llms.txt: the published evidence is a null.
2. Verify in Google Search Console and Bing Webmaster Tools; enable IndexNow (see the Bing playbook).
3. Publish 5–10 genuinely useful, stat-dense cornerstone pages, each answer-first (see how to get cited).
4. Run the technical GEO audit so nothing blocks crawling or rendering.
5. Ship one original-data asset in month one — the single highest-leverage thing you can publish.
6. Seed 3–5 brand mentions immediately, and set a 13-week refresh calendar.
7. Stand up the weekly dashboard above, and watch your logs for the first AI-bot crawl.