Every week someone asks me to "do GEO" as if it were a setting they forgot to switch on. There isn't one. After thirteen years building marketing for the Saudi market, I can tell you the vendors selling an AI-citation button are selling air. Here is what actually moves the needle: publish the clearest verifiable answer to a specific question, make the business and the author unambiguous, keep the page crawlable, and earn corroboration beyond your own site. No markup tag and no submission form guarantees inclusion. GEO is just evidence-led content and technical clarity, pointed at the surfaces that generate answers.
What AI systems actually cite
Answer engines do not cite brands; they cite passages. ChatGPT, Gemini, Perplexity, and Google's AI Overviews assemble a reply from a handful of sources and attach links to the ones that supplied a specific, checkable claim. In practice they favour a direct answer to the exact question, a number with its method shown, a definition a non-expert can follow, and a source whose identity is obvious. Pages that bury the answer under positioning language get skipped, and a claim an engine cannot attribute to a named, consistent entity rarely gets quoted.
So the unit of GEO is the citeable passage, not the homepage. A page can rank well in classic search and still never surface in an answer because no single paragraph stands on its own. Build for the paragraph an engine can lift without changing its meaning.
Choose a citeable question
Start narrower than “rank my brand in ChatGPT.” That phrasing is a wish, not a target. Pick one question a real buyer would type and pin down the details before you write a word:
- the exact buyer question;
- the market and language;
- the products, services, or entities involved;
- the answer you can support;
- the page that should own it;
- the prompts you will use for monitoring.
Questions with a factual, comparative, or procedural answer are easier to build for than broad brand prompts. A model can lift "what documents do I need for X" cleanly. It cannot lift "why is your company the best," and you should not want it to.
Build an answer worth lifting
A citeable section usually contains:
- a direct answer in the first sentences;
- conditions or exceptions;
- evidence or methodology;
- a useful next layer such as steps, criteria, or a calculation;
- a clear update date where freshness matters.
Cut the scene-setting introduction. Nobody is going to cite your warm-up. Write the answer so it stays accurate when a machine rips it out of the page and drops it next to four competitors.
Make the entity clear
The page should make it easy to identify:
- the organisation and its canonical name;
- the author or responsible expert;
- what the organisation actually does;
- the market it serves;
- supporting about, contact, service, and case-study pages;
- consistent names and claims across the site.
Structured data may reinforce this clarity. It does not replace useful content, reputation, or links.
Prove the entity, don't just declare it
Saying who you are is not the same as proving it. An answer engine trusts an entity it can corroborate: the same organisation name, the same author, and the same core claims repeating across your own site and across places you don't control. When those signals line up, a model can resolve "who is this" with confidence and grows far more willing to cite you.
Concrete proof carries more weight than assertion:
- an author with a real track record, credited by name on the pages they wrote;
- an about page and a contact page that agree with every profile that references you;
- case studies that show method and outcome rather than adjectives;
- third-party mentions that spell the entity the same way you do.
FIT Institute became citeable in Google's AI Overviews because the brand was described consistently as one specific education entity and backed by content that actually answered the course questions people asked. To see whether an engine has resolved your entity yet, watch how often it names you unprompted; the mechanics of that check are in how to measure AI search visibility.
Use evidence that can survive verification
Strong evidence includes:
- first-party data with scope and method;
- dated screenshots tied to a query and market;
- named examples with permission;
- original calculations with inputs shown;
- documented experience and limitations;
- reputable external sources where a claim is not yours.
Do not turn an observed result into a universal rule. If an AI citation appeared for one query in one locale, report exactly that and nothing more. I hold every number I publish to two readings: what the platform reported, and what was actually true on the ground. Most of the GEO case studies you read online never make that distinction, which is precisely why they read like fiction.
Give every claim a source page to land on
Every strong claim on a citeable page should point to a page that proves it. Answer engines and the buyers who follow the citation do the same thing: they click through to check. Proof that lives only in a slide you emailed a client once can be neither cited nor verified.
Build the pages that carry your evidence:
- case studies with the scope, the method, and the dated result;
- a methodology or "how we measure" page you can link to repeatedly;
- an author or about page that establishes who stands behind the claim;
- data or calculation pages where the inputs stay visible.
Report results the way they survive an audit. On one Saudi e-commerce program I reconciled 2.3M SAR of ad spend against 11.5M SAR of CRM-verified collected revenue — a clean 5.0× ROAS measured on money actually banked rather than the platform's attributed figure. Two numbers, both shown. The KSA reconciliation case exists precisely so that claim has somewhere to land, which is the whole point: a source page turns a nice sentence into something an engine, and a buyer, can trust.
Improve passage-level usefulness
Generated answers may rely on a small passage rather than the whole page. Help the passage stand alone:
- use descriptive headings;
- keep definitions concise;
- put labels next to numbers;
- explain acronyms;
- avoid references such as “as mentioned above”;
- use lists for steps and criteria;
- link to the underlying proof.
This also improves ordinary usability and featured-result potential.
Earn corroboration outside your site
Your own page can establish the answer; external signals help establish that the entity and claim are credible. Relevant industry mentions, partner pages, profiles, reviews, directories, interviews, and editorial links can all help when they are accurate and consistent.
Do not buy generic mentions at scale. Relevance and verifiability matter more than volume.
Technical checklist
- The page returns a normal successful response.
- Important content appears in rendered HTML.
- Search engines are allowed to crawl and index it.
- Canonical and hreflang signals are correct.
- Internal links point to it from relevant pages.
- The mobile page contains the same substantive answer.
- Titles and headings describe the question accurately.
- Structured data matches visible content.
- Evidence links work.
- Updates do not change the URL without a reason.
An llms.txt file can provide machine-readable guidance, but it does not cause citation and should never be the centre of the strategy.
Structured data and author schema
Schema markup describes your page to machines in a format they parse directly. It will not earn a citation on its own, and no engine has promised that it will. What it does is remove ambiguity: Organization and Person markup tie a page to a named entity, Article markup with an author and a dateModified signals who wrote the page and when it was last checked, and FAQ or HowTo markup labels the exact passages built to answer a question.
Two rules keep schema honest:
- it must match what a human sees on the page; marked-up claims that never appear in the visible content are a liability;
- author schema should resolve to a real, credited person with a track record rather than an empty byline.
Treat structured data as a clarifier for content that already deserves to be cited. It sharpens a strong page and can do nothing for a weak one.
Monitor citations as observations
Create a simple log with:
- engine and surface;
- exact prompt or query;
- language and market;
- date;
- cited URL;
- quoted or summarised passage;
- screenshot or saved evidence;
- whether the citation persisted on retest.
Test a stable prompt set over time. Do not search once, see your brand, and report a permanent visibility win.
Keep a refresh cadence
Citations decay. An engine that quoted you in March can quote a fresher competitor by June, especially on questions where the correct answer moves — prices, policies, model versions, "best of" lists. A page that earned a citation is a position you now have to hold.
Set a review rhythm by page type:
- fast-moving answers (pricing, tools, anything with a year in the title) — review every quarter;
- stable how-to and definition pages — twice a year;
- evergreen methodology and case pages — once a year, or whenever the underlying facts change.
When you refresh, change what is actually true first, then record it with a visible update date and an honest dateModified. Nudging a timestamp on a page you did not touch fools no reader and no crawler worth the name. Re-run your tracked prompts after each meaningful update so you can see whether the refresh held its citations or lost them.
A 30-day GEO sprint
Week 1: map
Choose 10–20 high-value buyer questions, assign one page owner to each, and record a baseline.
Week 2: improve
Rewrite the strongest opportunities with direct answers, evidence, limitations, and passage-level structure.
Week 3: connect
Fix internal links, entity information, crawlability, proof pages, and relevant external profiles.
Week 4: observe
Retest the prompt set, record citations, compare search performance, and plan the next update from evidence.
What success looks like
Success is not “being everywhere in AI.” It is:
- citations for commercially relevant questions;
- correct representation of the business;
- source URLs that buyers can visit;
- qualified traffic or enquiries from those pages;
- repeatable monitoring and improvement.
FIT Institute has been cited in Google’s AI Overview and ranked first for tracked course queries, including specific terms where it out-cited PwC Academy Middle East. That is useful observed proof, bounded to the tracked queries and captures. Read the case-led explanation in the FIT GEO teardown.
Next step
If you want this run for you — a query map, a citation baseline, entity and source-page fixes, and a refresh plan — request a GEO audit. For the broader commercial scope, see AI SEO and GEO.