GEO for New Websites: A 90-Day Plan to Become Visible to AI

A new website is invisible to AI assistants twice over. It is invisible the way it is invisible to Google: unindexed, unlinked, untrusted. And it is invisible a second way that is specific to assistants: the models have never seen your brand described anywhere, so even their training-data memory has nothing to say about you. We wrote the classic-search version of this playbook in SEO for a new website, and this article is its GEO twin: same 90-day discipline, same honest timelines, applied to AI visibility.

The plan has three phases that overlap on purpose: foundations in the first two weeks, content through day 60, authority from day 30 onward. The order is not negotiable, because each phase gates the next: assistants cannot cite what they cannot retrieve, and they do not choose what nothing vouches for.

The 90-day plan at a glance

Now each phase in working detail.

Days 1 to 14: foundations, or becoming retrievable

The direct answer: before any content strategy, a new site must be fetchable by three kinds of machine (Google's crawler, Bing's crawler, and the AI crawlers), and must present one unambiguous answer to the question "what is this brand?". Two weeks is enough for all of it.

Get indexed on Google AND Bing

Everyone sets up Google Search Console. Almost nobody on a new site verifies Bing Webmaster Tools, and for GEO that omission is expensive: ChatGPT's browsing draws on Bing's index, so a site absent from Bing is absent from a major assistant regardless of its Google status. Do both in week 1: verify the domain, submit the sitemap, request indexing for the homepage and key pages. Bing also supports IndexNow for near-instant discovery of new URLs; most CMS platforms have a plugin.

Open robots.txt to AI crawlers, deliberately

Assistants can only use what their crawlers may fetch. Decide explicitly, in week 1, which of these you allow: GPTBot and OAI-SearchBot (OpenAI documents both in its bot documentation), PerplexityBot (documented in Perplexity's crawler docs), and Google-Extended, which governs some Google AI training uses while normal Googlebot access covers search and AI features. The default state of many site templates and CDN configurations is to block some of these silently, so check the live robots.txt, not the one you think you deployed. For a site that wants AI visibility, the correct posture is open; blocking is a legitimate choice only when content protection outranks discovery.

Define your entity once, then repeat it everywhere

Models learn brands from repetition. Write the one-sentence description of what your company is ("X is a [category] that [does what] for [whom]") and deploy it verbatim: homepage, about page, footer, social profiles, directory listings. Then encode it in Organization schema on the homepage: name, url, logo, description, sameAs links to your profiles. A new brand has exactly one advantage here: no legacy inconsistency to clean up. The cost of getting this wrong compounds quietly; the mechanics of why assistants draw a blank on unknown brands are covered in why ChatGPT doesn't know your brand.

Publish llms.txt

Ten minutes: a markdown file at /llms.txt listing your canonical pages with one-line descriptions. Platform adoption is still partial, so treat it as cheap hygiene rather than a lever, the agent-facing equivalent of a sitemap. It costs nothing and occasionally helps an agent land on the right page.

By day 14, run the exit test: your key pages indexed on both engines (site: searches or the respective consoles), your live robots.txt permitting the crawlers you chose, your schema validating, and the same brand sentence everywhere a machine might read it. Everything in this phase, plus the checks we did not have room for, is itemized in the GEO audit checklist.

Days 15 to 60: content assistants can lift

The direct answer: a new site should spend these six weeks building one tight topic cluster of answer-first pages, not spreading across every topic it eventually wants. Assistants cite the source that most cleanly resolves a question, and a small site can only be that source for a narrow territory.

One cluster, chosen for winnability

Pick the single topic where your claim to authority is most defensible and the existing answers are weakest, and build 8 to 15 pages that resolve it completely: one hub page, satellites for every adjacent question, all interlinked. The keyword-selection discipline is identical to classic SEO for a new domain (winnable beats voluminous, every time), and the reasoning behind clustering is simple: twenty scattered pages teach the machines nothing, while twelve pages on one theme teach them exactly what you are the answer to. If the vocabulary here is new, the primer is what is generative engine optimization.

Answer-first, on every page

Structure every page so a machine can lift the answer without reading your life story: a question-shaped H1, the direct answer in the first paragraph, H2 sections that each resolve one sub-question and open with their own direct answer, comparison tables in real HTML, and an FAQ block with FAQPage schema targeting the People-Also-Ask variants. Google's own guidance on AI features says there is no special AI markup beyond normal best practices, which is the point: quotable structure is just ruthless clarity. How retrieval then turns those pages into citations is the pillar's territory, one paragraph of summary being enough here: assistants search, read the top results, and synthesize with citations, so structure decides what they can quote and rankings decide what they read. The full mechanics live in how to rank in AI search.

Ship one citable asset

Somewhere in these six weeks, build the one page that gives machines and humans a reason to reference a site with no reputation: a small original dataset (even 40 hand-collected data points beats zero), a free calculator or generator, or a genuinely reusable template. Assistants answering how-much and how-many questions prefer concrete, reproducible sources, and other sites link to assets far more readily than to opinions. This single page will do double duty in phase three.

We should say plainly where Meeeters sits in this phase, since content production is half of what we build: our audit proposes the cluster and the missing pages from your real site structure, and generates drafts answer-first with the FAQ and schema baked in, delivered to your CMS for you to review and publish. On a new site, that compresses the six weeks, but it does not skip the judgment: a human still decides what ships.

Days 30 to 90: authority, the gate nobody skips

The direct answer: for a new domain, backlinks and brand mentions gate AI citations more than anything else you do, because assistants choose sources that rank and brands that the web already describes, and a new site starts with neither. Phase three starts at day 30, overlapping content production, because links are the slowest variable and starting them "after the content is done" means starting months late.

The realistic day-90 target is modest: 5 to 15 referring domains from relevant, real sites, plus your brand described (in your boilerplate sentence, ideally) on a handful of third-party pages. Where those come from on a new site, in ascending order of effort:

• Existence links: relevant industry directories, professional profiles, local listings if applicable. Weak individually, but they seed the brand-mention graph that models learn entities from.
• Communities you genuinely participate in: useful answers in niche forums and groups, with occasional relevant links. Assistants increasingly retrieve from community discussions, which makes authentic participation quietly valuable twice.
• Your citable asset, promoted: the dataset or tool from phase two, offered to the newsletters and bloggers in your niche who cover such things. One asset outperforms months of cold outreach for a no-name domain.
• Guest posts and podcasts: one or two in 90 days is a fine pace, and each one adds a third-party page describing your brand in your own words.
• A link network done correctly: this is the Meeeters mechanism, and it exists precisely for the new-site cold start. You give an editorial link to one relevant site in the network, and a different site gives a verified link back to you, so there is no reciprocal pattern for anyone to detect or discount. For a domain with no audience and no budget for digital PR, it is one of the few ways to get relevant, editorial, dofollow links flowing in the window when they matter most.

Keep the velocity boring: a few links per month from varied sources. A week-old domain acquiring 50 links in a week looks exactly like what it usually is.

What day 90 honestly looks like

The direct answer: at day 90 a well-executed new site is retrievable, quotable and beginning to be vouched for, and it is still probably not being cited. That is the plan working, not failing.

A new domain rarely gets cited before it ranks, and new domains rank slowly; the assistants added a stricter jury on top of Google's, not a side door around it. Realistic day-90 numbers: pages indexed on both engines, impressions climbing in Search Console, 5 to 15 referring domains, and your first mentions when you manually ask assistants the ten questions you care about, logged monthly in a spreadsheet (the only tracking a new site needs; paid trackers earn their fee once the mention rate is not zero). First real citations typically arrive in months 4 to 8, on the long-tail questions your cluster owns, then compound: cited pages earn mentions, mentions feed the next model, and the flywheel that felt impossibly slow in month 2 starts feeling unfair in your favor by month 10.

The failure modes are the same five as classic SEO (buying spam links in month one, publishing thirty thin AI pages in week one, redesigning instead of publishing, changing URLs casually, quitting at week 10), and the fix is the same: run the phases in order and let the boring inputs compound.

A measurement routine that costs nothing

The direct answer: a new site needs exactly three measurements, all free, on a fixed cadence. Paid tracking earns its subscription later; discipline earns its keep now.

Weekly, ten minutes. Search Console performance report, last 28 days: are impressions higher than last week, and did any new query appear? Impressions are the heartbeat of a young site; clicks and citations are too rare to read weekly, so do not try.

Monthly, thirty minutes. The prompt log: ask ChatGPT, Perplexity and Google's AI features the same ten money questions, in a fresh session, and record three things per prompt: is your brand mentioned, is your site cited as a source, and who is cited instead. The "who instead" column is the most valuable data a new site can collect, because the cited page shows you concretely what authority and structure the assistant preferred. Read one cited competitor page per month and note what it has that yours lacks; the answer is almost always some mix of rankings, referring domains and cleaner structure.

Monthly, in the same sitting. Referral traffic from assistant domains in your analytics, plus your referring-domain count. Log all of it in one spreadsheet. Ninety days of that log will tell you which phase of the plan is lagging, which no dashboard can tell you about a site this young.

What you should refuse to do: check citations daily (the sample noise will wreck your morale), and compare your month 2 to a competitor's year 5. Slow is the design, not the defect.

Start with the baseline: what Meeeters hands a day-zero site

Everything in this plan flows from knowing what a machine sees when it looks at your site today, and a new site has one honest advantage: nothing to undo. Meeeters was built to run this exact sequence, foundations then content then authority, from one place, which is why most members start it in week 1 of the plan rather than after it:

• The free analysis maps day zero in minutes: it crawls the site, maps your structure and silos, detects the schema you published in the foundations phase, and lists the missing cluster pages and the quick wins. No card required.
• Phase-two drafts come from that audit, not from generic prompts: each article targets a gap in your own cluster, in your site's language, so the winnability discipline is enforced by the tooling instead of by willpower.
• Everything arrives in your CMS as a draft (native Webflow connector, or a webhook via Make, Zapier or n8n) and nothing auto-publishes; on a domain this young, that human filter is the whole quality moat.
• The link pool is vetted for exactly the caution phase three demands: real sites only, dofollow links, and casinos, adult and directory sites banned outright.
• The give-first credit system keeps velocity boring by design: give one verified link, the network owes you one back, a pace that matches the few-links-per-month target instead of the 50-in-a-week spike that marks a new domain.
• Matching runs on language and audience worldwide, so the 5 to 15 referring domains you target by day 90 come from sites whose readers can actually become your visitors, and the Google Search Console integration tracks the impressions heartbeat from the same dashboard.

There is a free plan to start and paid tiers on /pricing when the cadence needs to rise, but the plan begins the same way for every site: run the free SEO analysis and take an honest look at day zero.

Christopher Fernandes

Founder of Meeeters

I built Meeeters to make link building safe and simple: real, relevant backlinks with no reciprocal footprint and no black-hat shortcuts. Questions about your site? Write to me directly.

Email us

We reply fast, usually within a few hours.