Short answer, as of August 2026: crawler and indexing fixes can show up within days; a genuinely new page typically starts appearing in AI answers two to six weeks after it is published; and a stable, defensible share of answers for a competitive topic takes a quarter or more. If someone promises citations "this week" for a page that does not exist yet, they are describing a coincidence, not a process.
That said, the fastest-moving parts of this work are also the cheapest and the most commonly skipped. If you are in a hurry, do those first — in order.
Why AI citations lag, and what the lag is made of
An AI answer is assembled from an index, not from the live web. Between publishing a page and being quoted in an answer sit four separate delays, each with its own clock:
- Access. The retrieval crawler has to be allowed to fetch you at all. This gate is binary and it is the only one that can be fixed in an afternoon.
- Discovery and indexing. Something has to notice the page exists and store it. Days for a well-linked site with a working sitemap and IndexNow submission; weeks for a site with neither.
- Retrieval. For a given prompt, the engine has to select your page from the index as one of a handful worth reading. This is where relevance, specificity and entity clarity do their work.
- Selection. Having read it, the model has to decide your page answers the question well enough to quote. This is the part content quality actually controls.
Almost everyone measuring "how long GEO takes" is measuring all four at once and blaming stage four for a stage-one problem. Separating them is most of the diagnosis.
The realistic timeline, stage by stage
| Stage | Typical elapsed time | What you actually do | How you know it worked |
|---|---|---|---|
| Crawler access unblocked | Hours to 3 days | Allow retrieval agents in robots.txt; whitelist them at the WAF/CDN | Live requests with the crawler user-agent return 200 |
| Baseline measured | 1–3 days | Build a prompt set, run it across the engines, record who is cited today | A citation rate and share-of-voice number you can date |
| New/updated page indexed | 2 days – 3 weeks | Publish, submit via IndexNow and Bing WMT, link internally from an existing page | Page appears in index coverage; live fetch by search bots |
| First AI citations on the new page | 2–6 weeks from publish | Nothing — this is waiting, with re-checks on a schedule | The tracker flips one or more engines from "not cited" to "cited" |
| Measurable movement on the prompt set | 6–12 weeks | Iterate on the prompts you lose, fix entity inconsistencies | Citation rate across the set trends up run over run |
| Durable share of answers | 3–6 months | Depth of coverage, third-party corroboration, freshness upkeep | You hold citations through re-runs, not just once |
Two caveats keep that table honest. First, ranges are ranges: a well-established domain publishing on a topic it already owns can compress the middle rows dramatically, and a brand-new domain with no external references can blow past the top of every range. Second, the engines do not move together — it is normal to be cited by Perplexity weeks before ChatGPT, because their index refresh behaviour differs. Which is why measuring per-engine rather than as an average matters; we cover the mechanics in how ChatGPT, Perplexity and Gemini pick their sources.
If you need results immediately, do these four things
"Immediately" is the wrong frame for citation-building, but it is exactly the right frame for the failures that are preventing citations. All four of these can be done inside a week, and each one either works or does not — no waiting to find out.
1. Verify crawler access at the firewall, not just in robots.txt (one afternoon). This is the single most common silent blocker, and the fix is free. A permissive robots.txt means nothing if your CDN's bot-management ruleset returns 403 to unfamiliar user-agents. Request your top pages with each retrieval agent's user-agent and confirm a 200. The full audit is in the AI crawler access checklist.
2. Check that your content renders without JavaScript (one afternoon). If the answer text only exists after client-side hydration, several retrieval agents will read an empty shell. Fetch a page with JavaScript disabled and confirm the substance is in the HTML. Rendering and AI search visibility covers what to check and how to fix the common framework patterns.
3. Fix the pages you already have rather than commissioning new ones (a few days). An existing page with existing authority, restructured to answer a specific question directly in its first two sentences, will beat a brand-new page to a citation almost every time — it skips most of the discovery delay. Retrofitting existing pages for AI citation is the highest-yield week of work available to most sites in a hurry.
4. Make your entity data consistent (a few days of tedium). Name, address, phone, description and category should match across your site, your Google Business Profile and the major directories. Inconsistency does not just dilute local ranking; it makes an engine less confident that the three sources it found are the same organisation. See entity consistency and the knowledge graph.
Notice what is not on that list: publishing twenty new posts this week. Volume is a medium-term lever, and it is the one most likely to be recommended by whoever is billing hourly.
What genuinely cannot be rushed
Three things resist money entirely.
Index refresh cadence. You do not control how often an engine re-crawls or rebuilds. You can shorten discovery with sitemaps, internal links and IndexNow pings, but after that the ball is in their court.
Third-party corroboration. Engines lean heavily on sources that are not you — reviews, directories, forum threads, trade publications, comparison pages. Earning those takes as long as earning anything from other people. It is also the most durable advantage in the whole discipline, precisely because it cannot be bought overnight.
Track record on a topic. Depth reads as authority. One good page on a subject is a data point; twelve interlinked pages that cover the subject properly is a pattern, and patterns get selected more often. That is a quarter of work minimum, and no amount of budget compresses it below the publishing cadence you can sustain.
What to measure at each checkpoint
The reason GEO feels slow is usually that nobody set a baseline, so week six looks identical to week one. Fix that before you start.
Day 0 — baseline. Run your prompt set across every engine and record, per engine: were you cited, who else was, and in what position. That is your before picture. Do this even if the answer is "cited zero times out of sixty" — especially then, because zero is a perfectly good baseline and the first non-zero is unmistakable.
Week 2 — access and indexing. Not "are we cited yet" but "can we be". Crawlers returning 200, pages in the index, content present without JavaScript. If any of these is false, the citation clock has not started.
Week 6 — first movement. Expect the first flips on retrofitted pages, not new ones. Look at which prompts moved, not just the aggregate, because the pattern tells you what the engines liked.
Week 12 — trend. Now the aggregate means something. Citation rate across the set, share of voice against named competitors, and whether the gains hold across consecutive runs or flicker in and out.
A caution about that flicker: AI answers are non-deterministic. The same prompt asked twice can produce different sources. A single run is a sample, not a measurement — which is why re-running on a schedule and looking at the trend is the only defensible way to read the data. Our buyer's guide to AI visibility trackers goes into what to demand from any tool on this point.
The honest version of "how fast"
If your crawlers are blocked today, you could be measurably better inside a fortnight — because you were not competing at all, and now you are.
If your access is fine and your content is fine and you are simply not being selected, the timeline is months, and the work is depth and corroboration rather than speed.
Most sites are in the first category and assume they are in the second. That is the good news, and it is the reason the access audit comes before the content plan every time.
DigiRank's AI Visibility Tracker exists to make this measurable rather than anecdotal: it re-asks your prompt set across ChatGPT, Claude, Gemini, Perplexity, Copilot and Google AI Mode on a schedule, charts citation rate over time, and flags a crawl-access regression as an issue in the same timeline as the visibility drop it caused. Tracking starts on the $99/mo Starter plan, and the audit-plus-auto-fix pipeline that watches for access regressions is included from Agency at $249/mo with a 14-day trial. If you are running this for clients, the 30-day agency playbook sequences the first month.
Frequently asked questions
How long does it take to get cited by ChatGPT? For a newly published page, typically two to six weeks after it is indexed, assuming ChatGPT's retrieval agents can fetch your site. For an existing page you restructure to answer a specific question directly, it can be considerably faster because the discovery delay is already behind you.
Can I get AI search results immediately if it's urgent? No — but you can remove the blockers immediately, and that is often where the delay actually lives. Verify crawler access at your firewall, confirm your content renders without JavaScript, retrofit your strongest existing pages and fix entity inconsistencies. All four fit inside a week and none of them requires waiting for an index refresh.
Why am I cited by Perplexity but not ChatGPT? The engines use different indexes with different refresh behaviour and different retrieval agents, so appearing in one before another is normal rather than a fault. It is also why per-engine measurement is more useful than an averaged score — an average hides which engine actually moved.
Do I need new content, or can existing pages work? Existing pages usually work faster. They are already discovered, already linked and may already have external references, so restructuring one to answer a target question directly skips most of the delay a new page has to sit through.
Why do my citations appear and then disappear? AI answers are non-deterministic — the same prompt asked twice can surface different sources, so a single run is a sample rather than a measurement. Judge results on a trend across repeated scheduled runs; intermittent citation usually means you are a borderline candidate the engine sometimes selects, which is a signal to add depth rather than to start over.
Does publishing more posts speed this up? Volume helps in the medium term by building topical depth, but it does nothing for a blocked crawler or an unrenderable page — and those are the more common causes of a stalled result. Fix access first; publish second.
How often should I re-check my prompt set? Weekly while you are actively working on a topic, monthly once you are maintaining a stable baseline. Weekly gives you enough runs to distinguish a real trend from non-determinism within about six weeks.
When should I conclude it isn't working? Not before twelve weeks, and not before you have confirmed access, indexing and rendering are all clean. If those are verified, the prompt set has run weekly for a quarter, and the citation rate is flat, the problem is competitive depth rather than time — and the answer is more thorough coverage of fewer topics.