What to expect in the first 90 days with a GEO consultancy
The calendar we practise, week by week, and the three promises no consultancy controls in that window. A yardstick for any proposal.
First, the term, because the confusion is real: GEO here is Generative Engine Optimization, the work of getting a brand cited in the answers of AI assistants such as ChatGPT, Gemini or Perplexity. It is not geomarketing, it is not organisational strategy, it is not international expansion.
With that out of the way, the right question for whoever is hiring: what does a GEO consultancy deliver, week by week, in the first 90 days, and what can none of them promise in that window? This piece answers with the calendar we practise, to serve as a yardstick for any proposal.
Weeks 1 and 2: baseline and technical foundation
Nothing visible on the outside yet, and that is how it should be. The central deliverable is the measurement baseline: the category's buying questions run across the AI engines and it gets recorded where the brand appears, where it does not, and who appears in its place. In parallel, the short-cycle technical fixes are closed out:
- Schema.org on the key pages (Organization, Service, FAQPage), with the entity linked by
sameAs. robots.txtauthorising the relevant AI crawlers, andllms.txtpublished.- Key pages rewritten to open with the answer in the first sentence, in self-contained passages an engine can cite on their own.
By the end of week 2 you should have in hand: the opening photograph with numbers, the prioritised list of gaps and the plan for the quarter. If the proposal does not include a measured baseline, the rest of the calendar is not verifiable.
Weeks 3 to 6: content that answers, and entity
With the foundation closed, the work turns to what the engines cite. The typical deliveries of this block:
- Decision content: pages that answer the category's real buying questions (comparisons, criteria for choosing, costs, timelines), one question per section, in the buyer's words.
- Entity: consolidation on Wikidata, consistent profiles and descriptions across every platform, named authors with credentials on the key pages.
- Continuous measurement running: the same battery of questions repeats on a fixed cadence, and the report starts showing week-on-week variation rather than loose snapshots.
The scientific base for this phase comes from the Princeton work that gave the discipline its name: Aggarwal et al. (KDD 2024) measured that content and source optimisations can raise visibility in generated answers by up to around 40% under the study's conditions. It is the argument for working on citable content rather than waiting for luck, with the honest caveat that results per brand vary with the category and the competition.
Weeks 7 to 12: authority and reading the trend
The last block attacks what takes longest and pays most: the external sources the engines cite when they decide who to recommend.
- Presence in the category's sources the engines already cite: press, relevant directories and lists, communities where the category is discussed.
- Original data published (a study, a barometer, numbers of your own with a methodology), because citable data is the format that attracts citation most.
- Reading the trend: at the end of the quarter, the weekly measurement lets you compare before and after with the same questions and the same engines.
The public reference for the fundamentals is still Google's own documentation (Search Central): the AI functions of search rest on the same principles of useful, crawlable, verifiable content as the classic index. There is no technical shortcut that replaces this, and whoever sells one is selling risk.
What not to expect in 90 days
Three promises no consultancy controls in this window, and which serve as an honesty test for any proposal:
- A guaranteed position in an AI answer. The models do not sell positions; whoever guarantees one is inventing.
- Entry into the models' training memory. What the models know without searching the web changes in training cycles, measured in months. In 90 days you work the search route; the memory is background work that starts now and pays later.
- A Wikipedia article. Without prior independent press coverage there is no eligibility. The right order is press first, encyclopaedia after.
To close
A quarter well spent with a GEO consultancy ends with the technical foundation closed, continuous measurement running, decision content published and a legible trend in the numbers. Ask any supplier for this calendar in writing, ourselves included. The measurement that underpins each phase is Periscopy.
Frequently asked questions
When do the first GEO results appear?
It depends on the mechanism. In the engines that search the web in real time (Perplexity, ChatGPT with search, Google AI Overviews), well-structured new content can be cited within weeks. In the models' training memory, which answers when there is no web search, the change takes months and depends on presence in durable sources. An honest consultancy separates the two timelines in the plan; 90 days is enough for the foundation and the first citations in search mode, not for moving the models' memory.
What should be measured in the first month?
The complete baseline: in which buying questions the brand is cited, in which engines, how often, and who is cited in its place. Without that opening photograph, no later result is demonstrable. The measurement should have a fixed cadence and stable questions, so the weeks are comparable with one another.
What is impossible to promise in 90 days?
A guaranteed position in any AI answer, presence in the models' training memory, and a Wikipedia article without prior independent press coverage. Whoever promises those three things in that window is selling what they do not control. What is fair to demand: the technical foundation closed, measurement running, citable content published and a legible trend by the end of the quarter.
See also: how much a GEO consultancy costs in Portugal, how to choose one, and the two answer modes (why the timelines differ between search and memory).