What does a GEO company do? Deliverables, week by week

What does a GEO company actually do?

Generative engine optimization, or GEO, is a young service, so buyers often ask what they would be paying for. The honest answer is that a GEO company does a mix of research, technical work, writing and measurement. Most of it is quiet work on your website and your public profiles. Very little of it is visible from the outside until the end of the first month, when the before and after report arrives.

The work is built around one question. When a buyer asks an AI engine for what you sell, does the engine name you, and does it get the facts right? Everything a GEO company does is a way to improve that answer, or to prove that it did or did not change. If you want the longer definition first, read what is generative engine optimization.

This guide describes the work as it happens: a typical day, the deliverables week by week, what you need to do as the client, the parts nobody controls and the checks you can run yourself. We use our own GEO Pilot as the worked example, because it is the engagement we know best. Other companies will organise the work differently, and the section on checking the work applies to any of them.

One boundary matters from the start. A GEO company can change what you publish and how readable it is to AI systems. It cannot change what an engine decides to say. Any provider that talks as if it can is describing a different job.

What does a day in the life of a GEO engagement look like?

There is no single typical day, because the work moves through phases. A first week looks nothing like a third week. But the daily tasks fall into a small set of types, and it helps to know them so that a status update from your provider makes sense to you.

Here is a realistic map of a working day in each phase of a 30 day implementation. It is based on how we run the Pilot, so treat it as an example of the pattern and not a universal timetable.

Inside those phases, a day usually mixes four kinds of work. The first is observation: running prompts and recording what each engine says, with the date, the engine and the exact wording. The second is diagnosis: working out why a page is missing from an answer. The third is production: changing code, copy and markup. The fourth is communication: telling you what changed and asking for what we need.

Observation: reading what the engines say

Observation is slower than it sounds. AI answers vary from one run to the next, so a single screenshot is weak evidence. A careful provider records each prompt, each engine, the date and the full answer, then looks for patterns across several runs. The aim is a baseline that can be compared later, not a lucky capture.

Diagnosis: finding the reason a page is not used

When a competitor is named and you are not, the cause is usually one of a few things. The page may be blocked to a crawler. It may bury the answer under marketing copy. It may state a fact differently from your other pages and profiles. Or the engine may simply prefer a source that is more specific. Diagnosis means testing each of these rather than guessing. Our post on how AI search engines choose sources explains the usual factors.

Production: the changes themselves

Production is where a GEO company differs from a reporting tool. A tool tells you a page is weak. A company changes the page. That can mean editing a robots.txt file, correcting a canonical tag, writing a two sentence direct answer at the top of a service page or adding FAQ markup that matches visible text. All of it goes through your approval first.

Communication: short, dated, specific

Good updates are short. They say what was changed, on which URL, on which date, and what you need to decide. If your provider sends long updates with no URLs, you will find it hard to check the work later.

What are the deliverables, week by week?

Deliverables are the things you can open, read and keep. A useful test for any GEO engagement is to ask for the list of deliverables in writing before you pay. Vague promises such as "AI visibility improvements" cannot be checked. A file, a changed page or a dated report can.

Here is the sequence for the Pilot, which is a fixed $999 engagement delivered inside 30 days. It covers one domain and one language. A new site build, ongoing digital PR, link building and paid media are not included, and nothing in it is a promise of rankings, citations or traffic.

Notice what is missing. There is no deliverable called "ranking improvement" or "more citations", because those are outcomes, not deliverables. The report tells you what the engines said before and after. It does not claim credit for movement it cannot prove.

For longer engagements the deliverables repeat on a monthly cycle. Our pricing page sets out the plans. As of October 2026 the Growth plan starts from $1,500 per month and covers one domain, up to 40 tracked buyer prompts across five engines, 4 pages written or refreshed per month, a monthly technical check, entity and profile upkeep, a monthly report and a 30 minute call. The Scale plan starts from $3,000 per month and adds up to 100 tracked prompts, 10 pages per month, third party authority work and up to two markets or languages. Scope is confirmed in writing before any work starts.

What changes on your website?

Most buyers expect a GEO company to produce reports. The larger part of the job is changes to the site itself. They fall into three groups: access, structure and facts.

Access: can AI systems reach the page?

An AI engine cannot cite what it cannot fetch. Access work checks robots rules, server responses, sitemaps and page rendering. It also separates different kinds of bots, because they serve different purposes. OpenAI's documentation, which we checked in October 2026, describes OAI-SearchBot as the crawler used to surface sites in ChatGPT search, GPTBot as the crawler for content that may be used to train its models, and ChatGPT-User as the agent that visits pages when a user asks ChatGPT to. It also says that for search results it can take about 24 hours from a robots.txt update for its systems to adjust.

That distinction matters because blocking a training crawler and blocking a search crawler are different decisions. A GEO company should explain which bots you allow and why. Our guide to AI crawlers and what to allow covers the choices, and the free AI Crawler Access Checker lets you see your current state yourself.

Structure: can an engine find the answer quickly?

Structure work changes how a page is written. Each priority page gets a clear question it should answer, a direct answer near the top, headings that match how buyers phrase the question, and supporting detail below. FAQ blocks are added where real questions exist. Structured data is added only where it matches visible text.

Google's own documentation on AI features in Search, last updated 10 December 2025, says there are no additional requirements to appear in AI Overviews or AI Mode and no special optimizations needed beyond being indexed and eligible to show with a snippet. It also recommends that structured data match the visible text of the page. That is a useful guard against any provider who claims a secret markup trick. Structure work is about clarity for readers first, and machines second. See schema markup for AI search for the detail.

Facts: does the web agree about who you are?

Entity work compares the facts about your brand across your site, your social profiles, your business listings and other public pages. Names, addresses, services, founding dates and descriptions should agree. When they conflict, an engine may repeat the wrong one. The fix is mostly editing and tidying, and it is covered in entity SEO for brand consistency.

What does the client do during a GEO engagement?

A GEO company cannot work alone, and the engagements that go badly are usually the ones where the client was absent. The time you need to give is modest but it is real, and it falls on specific days.

• Give access in the first days: an admin login, a staging site or a Git repository, so changes can be made and tested.
• Name one decision maker who can approve changes within a day or two. Delays in approval are the most common reason a 30 day plan slips.
• Confirm facts we cannot verify: prices, service areas, credentials, legal wording, product details.
• Review page drafts for accuracy and tone. In regulated fields your compliance owner reviews wording before publishing.
• Join the closing call and read the report.

In our experience as a small agency, the time cost is a few hours spread across the month for a small business, and more for a company with legal review. That figure is our estimate, not a measured industry number. If your approval process takes three weeks, say so at the start so the plan can be built around it.

There are also things you should not do. Do not change the pages being rewritten during the month without telling the team, because it muddies the before and after comparison. Do not add new redirects or move the site halfway through. And do not ask for results in the first week. The baseline in week 1 is a measurement, not a result.

What can a GEO company not control?

This section is the one to read twice. A GEO company works on the inputs you own. The outputs belong to the engines, and the engines change often.

• Whether an engine names you in a given answer. Answers vary by run, by user, by location and by the engine's own updates.
• How an engine weighs sources. Each engine has its own retrieval and ranking, and none publishes the full recipe.
• Whether a particular crawler visits when you want. Documentation says what a bot is for, not when it will return.
• What other sites say about you. You can ask for corrections and publish better information, but third parties decide what they print.
• How fast changes show up. OpenAI says about 24 hours for its search crawler to adjust to a robots.txt update. Other systems give no figure.
• Whether AI answers send you traffic. Many answers are read without a click.

Because of these limits, a trustworthy GEO company talks in terms of preparation and evidence. It will say: here is what we changed, here is what the same prompts returned before and after, and here is what we still do not know. It will not say: you will be number one in ChatGPT.

This is also why we never promise rankings, citations or traffic. Our proof is limited to two real cases. Emtech, a Dubai IT company selling IT AMC, is now cited for IT AMC queries in Google AI Mode, ChatGPT, Microsoft Copilot, Perplexity and Gemini, and the work is described in the Emtech case study. Sudarshan Stoneix, a marble and granite company, saw in six months 1.48M Google impressions, 17.1K clicks, an average position of 4.9 and 909 AI mentions tracked by a third party tool, described in the Sudarshan Stoneix case study. Each is one client in one market. Neither is a promise for your site.

How can you check the work a GEO company did?

You do not need to be technical to verify most of a GEO engagement. You need the provider's records and about an hour. The checks below work for any company, including us. If a provider objects to any of them, that is information.

1. Ask for the baseline file. It should list each prompt, each engine, the date and the full answer or a faithful record of it.
2. Run five of the prompts yourself in a clean session. Answers will differ in wording, but you should recognise the same pattern of who is named.
3. Open the change list and visit three of the changed URLs. Confirm the change is live, not just described.
4. Run the AI Crawler Access Checker on your domain and compare it with the crawler finding in the report.
5. View the source of a rewritten page. The structured data should describe things a visitor can see on the page.
6. Read a rewritten page as a customer. Is the answer clear in the first lines, and is every claim true?
7. Ask for the retest. It must use the same prompts and the same engines as the baseline, or the comparison is not fair.
8. Ask what did not work. A report with no failures and no unknowns is a sales document.

Your own AI Visibility Checker result is a useful cross check, but treat any single tool score as one input. Our post on how to measure GEO performance explains what a fair measurement looks like, and the GEO reporting template for clients shows what a clear report contains.

Signs the work is real

• Every deliverable has a URL, a file or a date attached.
• The report states its method and its limits.
• Changes are documented well enough that another provider could take over.
• You keep everything: access, files, copy and reports.

Signs to be careful

• Results are described but not shown.
• Success is defined after the fact.
• The same screenshot appears in several places.
• Pricing depends on how many AI mentions are reported, or the provider guarantees an outcome.

For a fuller set of questions to ask before you hire, see how to choose a GEO agency. It covers red flags, proof and pricing models for any provider.

Example, hypothetical: one month with a small B2B services firm

This is an invented example to show the rhythm, not a real client and not a result. Imagine a 15 person IT services firm in Dubai that sells managed support contracts. Its buyers ask AI engines questions such as "who offers managed IT support for small companies in Dubai" and "what should an IT support contract include".

In week 1 the team runs 20 such prompts on five engines. The firm is named in a few answers and absent in most. The crawler check finds that a security plugin is returning errors to some AI bots. The entity audit finds the company describes itself three different ways across its site, its business listing and its social profile.

In week 2 the plugin rule is corrected, the sitemap is cleaned up, canonical tags on two service pages are fixed and organisation markup is added that matches the visible about page. In week 3 the six most important service and guide pages are rewritten so each opens with a direct answer to a real buyer question, with an FAQ block drawn from the sales team's most common questions. The owner approves the drafts, and corrects two prices.

In week 4 the same 20 prompts are run again. Perhaps the firm is named in more answers, perhaps in the same number. The report says which, shows the answers side by side and sets out a 90 day roadmap. If nothing moved, the report says so and explains which of the likely causes it checked. That honesty is the product.

What happens after the first 30 days?

A first month establishes a baseline, fixes the largest blockers and rewrites the most important pages. It does not finish the job. AI engines update, competitors publish and your own business changes. After the Pilot you have three reasonable paths.

None of these is the right answer for everyone. If your site is small and your team has capacity, the roadmap may be enough. If you sell high value services and need steady output across several pages every month, a monthly plan is usually easier to manage. For larger groups with several teams or regions, see enterprise GEO. If you are still deciding between providers, GEO Pilot versus a monthly retainer compares the two models fairly.

What to do next

Start with evidence. Run the free AI Visibility Checker and the AI Crawler Access Checker on your own domain. Write down ten questions a buyer would ask about what you sell and try them in two or three engines. That gives you a rough picture before you speak to anyone.

If you want us to look at it, ask for the free AI visibility report on the contact page. We prepare it by hand and reply within one working day. If you decide to go further, the GEO Pilot is $999, fixed, delivered inside 30 days and done for you. We run a small number of Pilots at a time so each one gets senior attention.