The 90-Day Rollout

Most GEO programmes fail on sequencing rather than effort. The technical fixes, the content rewrites, the community work and the measurement all interact: some cannot start until others finish, and some must start in week one even though the payoff arrives in month three. This plan names those dependencies and builds the order around them.

Why sequencing matters more than effort

The five parts before this one are not a menu. They have a load-bearing order, and running them out of sequence wastes the work. Schema on a page a crawler cannot reach does nothing. An account registered in month two cannot participate credibly in month three. A measurement programme started after the content changed cannot tell you whether the content changed anything.

The common version of this failure looks tidy: technical in month one, content in month two, community in month three. That plan is wrong in two places. Baseline measurement and account warm-up both belong in week one regardless of when their output gets used, and deferring them to the month they pay off in costs six to eight weeks that cannot be recovered.

This part sequences the work from Parts 1 to 5 and does not re-teach it. Where a step needs detail, the part that covers it is named, and you should go back to it.

The threads that run the whole ninety days

• Measurement runs every week from day one. The baseline is recorded before anything changes, and weekly runs continue from there. A programme that starts measuring in month three has no baseline and no trend, so the day-90 checkpoint produces one number with nothing to compare it to. Part 5 has the method.
• Account warm-up starts in week one. Part 4 puts the working requirement at a month of account age and a comment history across more than one subreddit, with four to six weeks of ordinary activity a reasonable budget before the account goes near your category. That cannot be compressed. Register in week one, use it normally through months one and two, and stay out of your category until month three.
• The change log runs from week one. Every fix, schema addition and rewrite needs a date in the tracking sheet. This is the most commonly skipped item on the list and the cheapest to maintain.

Week one: record the baseline and confirm access

Nothing on the site changes in week one. The whole week is diagnostic, because any change made before the baseline exists makes the baseline meaningless.

What week one produces

• A prompt set of 40 to 50 questions built against the criteria in Part 5: what buyers ask an AI engine when evaluating a purchase, not the keywords you rank for in Google.
• A baseline run of that set across all four engines. Part 5 is explicit that a single run per prompt is one draw from a non-deterministic system rather than a measurement, so the baseline is run at least five times per prompt, spread across different times of day. This is the largest single time cost in the plan and it is not the place to economise: every later comparison is made against this number.
• A crawler access audit. Fetch five to ten key pages with curl using each retrieval user-agent from Part 3, confirm a 200, and confirm the response body contains the actual content rather than a JavaScript shell. Record what fails.
• A Reddit account registered and posting its first ordinary comments in subreddits unrelated to your category.

Who does week one

The prompt set needs someone who knows your buyers: a product marketer, a content lead, a senior account manager. The crawler audit needs someone comfortable reading an HTTP response. The account needs a named person with a real professional background in the subject, because that person will be posting under their own identity in month three.

As a working estimate rather than a measured figure: four to six hours for the prompt set, eight to twelve for the baseline runs, two to three for the crawler audit, one for account setup.

Month one: the technical layer

Month one is Part 3: crawler access, rendering, schema, entity signals, crawl hygiene. These have the shortest path from work to effect in the whole programme, because they change what engines can see rather than what they find when they look.

The order inside month one

1. Crawler access first. Correct any robots.txt rule blocking OAI-SearchBot, Claude-SearchBot, PerplexityBot or Googlebot, and check the edge as well, since CDN bot rules are enforced before robots.txt is read. A blocked crawler makes every later fix invisible. Verify with curl -A "OAI-SearchBot" -I https://yourdomain.com/key-page after each change.
2. Rendering second. Find pages whose key content only exists after JavaScript runs. Part 3's position is that no vendor documents a rendering guarantee for its retrieval agent, so the safe assumption for OAI-SearchBot, Claude-SearchBot and PerplexityBot is that they do not see it. Move the content server-side or pre-render, and test by comparing curl output against the browser.
3. Schema third. Add or correct Organization, Article and FAQPage on priority pages, connected with @id rather than left as isolated blocks. Validate with the Schema Markup Validator, not Google's Rich Results Test, which dropped FAQPage when the rich result was deprecated in May 2026.
4. Entity signals fourth. Confirm sameAs covers Wikidata, LinkedIn and Crunchbase, and that the name, canonical URL and description match across all of them. This step gets skipped because it is not a code change. It is the step that tells an engine who you are when the brand name is ambiguous.
5. Crawl hygiene fifth. Redirect chains, broken internal links, canonical mismatches, sitemap accuracy, and the failures from the week-one audit.

Who does month one

Steps one to four need a developer and someone who can write JSON-LD. Step five can be run by a content or SEO lead with CMS and Search Console access. Working estimate: eight to twelve developer hours across the month, three to five for schema authoring, two to three for hygiene.

Month two: prompts and content

Month two is Parts 1 and 2: work out which pages the prompt set says matter, then rewrite them answer-first. The prompt set already exists. Month two is the first time it drives a decision.

Map prompts to pages

For each prompt, record which page on your site is the best available answer. Some prompts will have no good match, because the page does not exist or exists and is not structured for extraction. Those gaps are the month-two work.

Prioritise by buying intent rather than by how easy the rewrite looks. A late-stage comparison question is worth more than an awareness question even when the page behind it is harder to fix.

Rewrite the priority pages

Apply Part 2 to each page in priority order. Three elements are non-negotiable:

• A 40 to 60 word opening block answering the page's primary question, before anything else.
• H2 and H3 headings phrased as the questions a buyer types, not as category labels.
• A visible question and answer section covering the three to five questions the prompt set shows buyers asking, with matching FAQPage markup.

Do not rewrite everything. A rewrite not connected to a prompt in the tracking set cannot be measured, so it can wait.

Who does month two

Mapping needs whoever built the prompt set, because they know what each prompt is actually asking. The rewrites need a writer who can hold the answer-first structure without sliding back into brand voice. Working estimate: four to six hours for mapping, three to five per page. Four to six pages is a realistic month.

Month three: off-site and the first real read

Month three is Part 4, plus the first full measurement cycle against the week-one baseline.

Starting community participation

The account registered in week one now has around eight weeks of ordinary activity, comfortably past the four to six week budget in Part 4. The subreddits identified during month two's mapping are the starting point, and Part 4's model applies unchanged: answer existing questions, post only when you have something the thread does not already contain, and disclose the affiliation whenever you reference your own organisation's work.

Two to four substantive answers a week is a realistic rate for one person. Substantive means useful standing alone, specific to the question asked, and carrying no link unless the link is the answer. Track which threads you answered and how the comments scored, because that is the signal the community found them useful, and Part 4 explains why that is also the signal engines act on.

The first full measurement cycle

At the end of month three, run the full prompt set five times per prompt across all four engines, on separate days, then calculate presence rate and citation count per engine against the week-one baseline.

What this can honestly tell you: whether presence rate moved, on which engines, and which pages now appear that did not at baseline. What it cannot tell you: whether your work caused it, whether it will hold, or whether the answers citing you are recommending you.

What runs continuously, and why phasing it breaks the plan

The change log is the one teams skip. It costs under five minutes per change and it is the difference between a day-90 checkpoint you can interpret and a number with no story attached.

Who does what

Three distinct skill sets. Not necessarily three people, but each has to be named before month one begins.

Technical

Crawler access, rendering, schema, entity signals, crawl hygiene. Needs CMS access, server or hosting access, and the ability to write and validate JSON-LD. Does not need to understand GEO strategy, but does need to follow Part 3 without improvising. Working estimate: eight to twelve hours in month one, two to four in month two, one to two a month after.

Content

The prompt set, the mapping, the rewrites, the question and answer sections. Cannot be filled by someone without subject knowledge of the category, because a writer who does not understand the product cannot write an extractable answer to a buying question. Working estimate: four to six hours for the prompt set, eight to twelve for baseline runs, sixteen to thirty for rewrites, then a few hours a month for maintenance and weekly runs.

Community

Warm-up from week one, subreddit research in month two, participation from month three. Per Part 4 this cannot be automated, cannot be handed to someone without genuine knowledge, and cannot be shared across people on one account. It needs a named individual willing to post under their own identity. Working estimate: one to two hours a week during warm-up, three to five a week once participating.

The three checkpoints, and what each can honestly tell you

Each checkpoint is a question rather than a target: given what we have done, what should be visible by now, and what does it mean if it is not?

Day 30

The technical layer should be done. The question: do all key pages return 200 to all four retrieval agents, with the content in the response body?

If not, month two cannot deliver its full effect, because rewrites on pages that are still blocked or unrendered will not reach the engines they were written for.

What day 30 cannot tell you: anything about citations. Thirty days is not enough for changes to be crawled and reflected in responses. A flat presence rate at day 30 is the expected result, not a warning sign, and treating it as one is how programmes get abandoned a month before they could have shown anything.

Day 60

The first rewrites should be live and the account should be roughly eight weeks old. The question: are the rewritten pages appearing in responses where they did not at baseline?

Run the subset of prompts mapped to rewritten pages. If none appear, there are three candidate explanations and they have different fixes: the pages have not been recrawled, the rewrite did not produce a block meeting Part 2's criteria, or the pages are losing to sources with stronger authority signals.

What day 60 cannot tell you: how much month three will add, or whether month one or month two drove anything you do see.

Day 90

The first full cycle: the complete prompt set, five runs per prompt, spread across separate days, compared against the week-one baseline.

A rise in presence rate on one or more engines, alongside a change log showing what happened in the weeks before, supports the hypothesis that the work is doing something. It does not prove it. Engine behaviour, competitor content and phrasing drift all move presence rate on their own.

If day 90 is flat: do not change the prompt set and do not add pages to the rewrite queue. Audit in sequence instead. Crawler access first, since it may have regressed. Then extractability: does the curl response body carry a clear answer block near the top of each rewritten page? Then authority signals. Fix the first failure you find before looking at the next.

The full thirteen-week timeline

Owners: T technical, C content, R community.

What happens after ninety days

The technical layer moves to maintenance rather than rebuilding. Schema needs updating when pages change, hygiene needs a quarterly pass, entity signals need checking when a brand name changes or a product launches. An hour or two a month, not a project.

The content layer compounds. Each rewritten page produces measurement data that informs the next one, and the weekly prompt runs surface questions buyers are asking that were not in the original set. Over time the set becomes a live map of what your buyers ask AI engines, which is a more direct signal than a keyword tool measuring search volume.

The community layer is slowest to build and hardest to displace. An account with a year of substantive participation has standing a new account cannot buy or rush, which is precisely why Part 4 spends so long on what not to do.

And the measurement programme is what tells you which of the three is actually moving, and where the next unit of effort belongs. Without it you have maintenance without direction. With it, each ninety-day cycle gives you a sharper picture of what works in your category, against your competitors, for the questions your buyers actually ask. That picture is what the next ninety days run on.