Ninety days is enough for one honest cycle: set a baseline, fix access, improve the pages that matter most, publish evidence only you can offer, then review and decide. Keep the scope to one offer and a handful of pages. Google says results can take from hours to months to show, so plan the review for weeks after the work, and refuse the shortcuts that its policies treat as spam.
Choose a bounded scope
The plan fails when it covers everything. Choose one offer, one audience and ten questions: the ones that, if answered well, would bring the enquiries you want. Pick five to ten pages that should answer them, mixing existing pages to improve with a few to create. Decide who owns each page, how many hours a week the work will get and how you will know if it worked. The numbers should be small enough that a part-time owner can finish them.
Days 1–14: establish the baseline
Do not change content yet. First record where you stand, and remove access problems that make every later step less reliable.
| Task | Chapter | Output |
|---|---|---|
| Verify the site in Search Console (domain property) and Bing Webmaster Tools. Note the Generative AI performance report.[7] | Measure | Access and a baseline of impressions, clicks and AI impressions for the priority pages. |
| Check that key URLs return 200, robots.txt and canonicals are right and the sitemap is clean. | Crawl and index | A list of blocking issues and fixes. |
| Confirm the initial HTML of key pages contains the content. | JavaScript | A pass or fail per page. |
| Write the crawler policy and test it through every layer. | Crawler controls | A policy record and a tested robots.txt. |
| Write the identity record and fix what you control. | Brand consistency | One record, and a list of external corrections to request. |
| Build the intent map for the ten questions. | Search intent | Question, job, owning page, gap. |
| Run the first citation audit, five runs per question. | Citation audits | A baseline log and summary. |
| Add “How did you find us?” to the enquiry form. | Measure | A source field on every new enquiry. |
Days 15–30: improve the highest-value answers
Now work on the three to five pages closest to enquiries. For each, write a brief, then apply what the earlier chapters cover. Start with the answer at the top and its conditions, make the title and the first paragraph work as a preview, add contextual links to the next step and place the doubts that stop people beside the places where they arise. Fix the technical problems from phase one on the same pages. Date every change, and add an annotation to your Search Console charts.[2]
- Answer: answer-first sections and a short summary at the top.
- Preview: distinct titles and descriptions, and links with descriptive anchors.
- Decision: the objections handled where they appear, as in the FAQ chapter, and any comparison that fits.
- Experience: the page loads quickly and works on a phone. See page experience.
Keep the changes honest and modest. Rewriting to answer the reader’s question is exactly what Google’s guidance asks for, and nothing in it requires a special format for AI features.[3]
Days 31–60: publish evidence and connect the collection
With the key pages improved, add what only you can offer. Choose two or three pieces of original material from your real work: a documented project, a worked example, a method note, a decision comparison. Write them from the brief, with the evidence plan completed first, and apply the checks from evidence and authorship before publishing. Give each an author page and a review date.
Then connect them. Build or fix the hub, link each new piece from the pages where its question arises and check for orphans, as the architecture chapter describes. Ask the owners of outdated directory entries to correct them. Where AI helped with drafting, follow the evidence handoff and disclose it where readers would expect it.[4] Around day 45, run the audit again on a subset of the questions to check for obvious problems, without expecting changes yet.
Days 61–90: review, correct and decide
Rerun the full citation audit with the same questions, settings and run counts as the baseline. Pull the Search Console and Generative AI reports for the same equal periods, and compare organic and AI referral sessions, next-step clicks and enquiries with how they said they found you. Read the results as a pattern and note what would be surprising. Check the changed pages against pages you left alone. Correct any page that turned out to be wrong, and update the identity record and the crawler policy if anything has changed.
| If you see… | Then… |
|---|---|
| Impressions and next-step clicks up on changed pages, relative to the unchanged ones | Keep the pattern and apply it to the next five pages. |
| Impressions up, clicks flat, enquiries steady | Treat visibility as improved and keep measuring outcomes. Do not call it a win or a loss. |
| No change in anything, and the pages were already indexed and sound | Re-examine the question set and the intent match before writing more. |
| A page moved the wrong way | Check recent changes and annotations. Revert or fix only what the evidence supports. |
| Audit citations differ by less than the run-to-run swing | Do not report a change. |
Write the review as a one-page note and set the next cycle. A sensible long-term rhythm is a quarterly audit, a monthly measurement review and a yearly check of every page’s evidence and dates.
What to expect, and what not to do
Google’s starter guide says changes can take from a few hours to several months to show in search results, and suggests waiting a few weeks before judging. It adds that not every change will have a noticeable impact.[1] A 90-day plan therefore fits one cycle, and the review at the end should be as ready to find nothing as to find something. Set expectations with whoever is paying for the work.
| Shortcut | Why not |
|---|---|
| Generating many near-identical pages | Scaled content abuse and doorway abuse are named in Google’s spam policies, however the pages are made.[5] |
| Stuffing keywords into titles, headings and anchors | Keyword stuffing is a listed violation, and the GEO research found it did not help.[5][6] |
| Buying mentions or fake reviews | Google says inauthentic mentions are not as helpful as they seem.[3] |
| Adding llms.txt or special files as a fix | Google says Search ignores them.[3] The crawler documentation from OpenAI, Anthropic and Perplexity that we reviewed does not mention them. |
| Paying for a guaranteed AI ranking | No third-party tool has access to Google’s internal ranking or AI systems, and AI answers vary from run to run.[3] |
Apply it to your website
- You have chosen one offer, ten questions and five to ten pages, each with an owner and an hours-per-week budget.
- Days 1 to 14 produce a recorded baseline and a clean technical and access picture before any content changes.
- Days 15 to 30 improve the pages nearest to enquiries, with dated changes and chart annotations.
- Days 31 to 60 add two or three pieces of original evidence, linked into the collection and credited to real authors.
- Days 61 to 90 rerun the same audit and reports, compare changed pages with unchanged ones and end with a decision.
- Decision rules were written at the start, before the data arrived.
- You have refused the shortcuts that Google’s policies or the evidence say do not work.
Sources and further reading
Sources were consulted in October 2026 and the numbered notes above point here. Platform documentation changes often, so check the linked page before you act on a platform-specific detail. Worked scenarios are illustrative unless identified otherwise; methods and worksheets are Vantle’s editorial guidance.
- Google Search Central: SEO Starter Guide, “How long until I see impact”Changes can take hours to months; wait a few weeks; not every change has a noticeable impact.
- Search Console Help: Search Console annotationsMarking events on report charts.
- Google Search Central: Optimizing your website for generative AI featuresWhat to focus on and what to ignore, and third-party tools.
- Google Search Central: Google Search’s guidance on using generative AI contentAI-assisted content must be reviewed and must not be used for scaled content abuse.
- Google Search Central: Spam policies for Google web searchScaled content abuse, doorway abuse and keyword stuffing.
- Aggarwal et al.: GEO: Generative Engine Optimization (KDD 2024)The study in which keyword stuffing did not improve visibility.
- Search Console Help: Generative AI performance report (Search)Impressions in AI Overviews and AI Mode, available to all sites since 31 August 2026.