An AI SEO framework is a repeatable method for getting a brand cited inside AI-generated answers rather than ranked in a list of links. The good ones all cover the same four jobs. The names differ, the sequence differs, and the marketing differs enormously.
This page covers what the research actually proves, which frameworks exist, and the questions worth asking before you buy one. Our own framework is in here too, as one entry among several, because a page that only promotes its author is not worth reading or linking to.
Why frameworks appeared at all
Traditional SEO had a shared scoreboard. You ranked at a position, for a query, and the position was checkable by anyone.
Generative search removed the scoreboard. An assistant answers in prose, names a handful of brands, and attributes some of them. There is no position to report. Two people asking the same question can get different answers, and the same question asked twice can return a different set of brands.
Frameworks filled that gap. They are an attempt to make the work repeatable when the outcome is harder to see.
What the research actually proves
The foundational work is GEO: Generative Engine Optimization by Pranjal Aggarwal, Vishvak Murahari, Tanmay Rajpurohit, Ashwin Kalyan, Karthik Narasimhan and Ameet Deshpande, accepted to KDD 2024. It introduced GEO-bench, a benchmark of user queries across multiple domains, and tested content modifications to see which ones increased visibility inside generated answers.
Its headline finding: optimisation methods can boost visibility up to 40%.
“Up to” is doing heavy lifting in that sentence, and almost every page quoting it drops the phrase. It is an upper bound for particular methods on particular query types. It is not an average, and it is not a forecast for your site. If a proposal quotes you a flat 40%, the person writing it has read a summary rather than the paper.
What the paper does establish, and what matters more than the number: content changes alone move citation visibility. Adding statistics, quotations and cited sources to a page measurably improved how often generative engines drew from it. That is the empirical basis the whole category rests on, and it is why every credible framework has a content-structure component.
The frameworks that exist
A survey rather than a ranking, because they are mostly not competing on the same axis.
HubSpot, six stages
Discover, Architect, Integrate, Automate, Activate, Compound. Built around their own toolset, which is both its strength and its limit: the stages assume you are operating inside HubSpot. The sequence itself is sound and portable if you are not.
Hostinger, six tactics
A content-engineering playbook. Notable because they published their own result rather than a client’s: a 52% rise in their AI citation share across three months after reworking 100 pieces of content. That is their own reported figure on their own property, which makes it more checkable than most agency claims and worth more than a case study you cannot inspect.
The discipline split: GEO, AEO, AIO and SXO
Less a framework than a map of the territory, and widely used as one.
| What it targets | |
|---|---|
| GEO | Citation inside a generated answer |
| AEO | The direct answer slot: AI Overviews, featured snippets, voice responses |
| AIO | Using AI safely inside your own content and research operation |
| SXO | Whether the attention converts once the click or citation happens |
They overlap heavily. Treat an agency selling them as four separate retainers with scepticism: that is a pricing structure rather than a delivery one.
MarGen, the Synaptic Authority Engine
Ours. A multi-phase methodology sequenced as six steps: AI visibility audit, prompt cluster research, entity authority building, citation-optimised content, technical infrastructure, then measurement and iteration. It is published in full at the Synaptic Authority Engine rather than held back, because a reseller or an in-house team needs to be able to explain what they are buying without waiting on us.
We are not claiming it does work the others do not. It organises the same underlying jobs around regulated UK sectors, where claims have to survive compliance review.
What they all have in common
Strip the naming and every serious framework covers four jobs. If one you are evaluating misses a job entirely, that is the question to ask about.
1. Machine-readable identity. Organization schema, consistent entity data, sameAs links to the
profiles that corroborate you. A model with no entity to attach trust to has nothing to hang the
rest on.
2. Extractable content. Answer-first structure, question-shaped headings, FAQ markup where it genuinely matches the page. The test is whether a clean, quotable, attributable answer can be lifted out of your page without rewriting it.
3. Third-party corroboration. The off-site half. Models cross-reference before recommending, so citation depends on sources the model already trusts saying the same thing about you. This is the slowest component and the one most frameworks underweight because it is the hardest to sell as a deliverable.
4. Measurement built on citation, not position. If the reporting still leads with rankings, the framework has been relabelled rather than rebuilt.
What a framework cannot do, on our own numbers
Publishing the limits is the part most vendor pages skip. Ours, from 90 days of our own Search Console data:
Page-one visibility does not mean clicks. We hold 6,482 page-one queries. They produced 218 clicks. 6,442 of those queries earned zero clicks between them, carrying 65,650 impressions. No framework fixes that, because nothing is broken: the answer is being given above the result.
A majority of that visibility is machines. Splitting those zero-click queries by shape, 59% of the impressions came from long, conversational, multi-clause strings, which is an assistant fanning out rather than a person typing. Useful, and invisible in any click report.
Authority still gates the outcome. Banding every query by how often it was searched, the terms almost nobody searches average position 22.0 for us, and the terms searched 30 or more times average 32.5. The more people search a term, the worse we rank for it. A framework does not overcome a domain authority gap on a head term, and any that implies otherwise is selling.
Five questions to ask before you buy one
Reusable regardless of who you buy from, including us.
- What does it measure? If the answer is rankings and traffic, it is an SEO retainer with new labels. Citation share across named platforms is the unit that matters.
- Is the method published? A framework kept as a black box cannot be explained by your team to your board, and cannot be taken in-house later. Both are reasonable things to want.
- What does it claim it cannot do? A framework with no stated limits has not met reality yet.
- Can they show it on their own site? Including what failed. An agency that has never applied its own method to its own domain is asking you to go first.
- What happens in month four? The structural wins land early. Ask what the work becomes once schema, entities and content structure are done, because that is where most of the money goes.
Where to go next
If you want the full method rather than the survey, the Synaptic Authority Engine is published step by step. If you want to know where you currently stand before choosing any framework, the free AI Visibility Score checks the machine-readable identity and extractability layers on your own site in about a minute.
For how the disciplines fit together, the complete guide to GEO, AEO, AIO and SXO goes deeper on each. For what the engines are actually doing with your pages, our AI search statistics carry the sourced figures.
Common questions
What is an AI SEO framework?
A repeatable method for getting a brand cited inside AI-generated answers rather than ranked in a list of links. Every serious one covers the same four jobs: establishing a machine-readable identity, structuring content so an answer can be lifted out of it, earning third-party corroboration, and measuring citation rather than position. The names differ. The jobs do not.
Does the research prove AI SEO works?
Partly, and the headline number is widely misquoted.
The foundational paper is GEO: Generative Engine Optimization by Aggarwal, Murahari, Rajpurohit, Kalyan, Narasimhan and Deshpande, accepted to KDD 2024. It reports visibility gains of up to 40% in generative engine responses.
Up to is the operative phrase. That’s an upper bound for particular methods on particular query types, not an average you should expect. Anyone quoting a flat 40% hasn’t read past the abstract.
Which AI SEO frameworks actually exist?
Several, from vendors and from practitioners. HubSpot publishes a six-stage AEO methodology running Discover, Architect, Integrate, Automate, Activate and Compound. Hostinger publishes a six-tactic content playbook and reports a 52% rise in its own AI citation share across three months after reworking 100 pieces of content. Most agency frameworks, MarGen’s included, organise the same underlying work differently rather than doing different work.
What is the difference between GEO, AEO, AIO and SXO?
GEO targets citation inside a generated answer. AEO targets the direct answer slot: AI Overviews, featured snippets, the response an assistant reads aloud.
AIO is the operational side, using AI safely inside your own content and research process. SXO is what happens after the click or the citation, whether the page converts the attention it earned.
They overlap heavily, which is why splitting them into separate retainers is usually a pricing decision rather than a delivery one.
How do I evaluate an AI SEO framework before buying it?
Ask four things.
What does it measure, and is that measure citation rather than ranking? Is the method published in full, or held back as a black box?
What does it claim cannot be done, because a framework with no stated limits hasn’t been tested against reality? And can the agency show the method applied to its own site, with the numbers, including the parts that didn’t work?
Can a framework guarantee AI citations?
No, and treat any that claims to as a warning. Citation depends on third-party corroboration accumulating on sources the models already trust, which is outside any agency’s direct control. A framework can make you eligible and can shorten the timeline. It cannot compel a model to quote you.
How long does an AI SEO framework take to show results?
Structural work can surface within weeks, because fixing schema, entity consistency and answer-first structure removes things actively blocking extraction. Citation authority typically takes three to six months, because it depends on other sites corroborating you. In our own published case studies the fastest meaningful movement was zero to 31% citation share across 45 queries in 90 days, and the longest ran 14 months.
Is an AI SEO framework different from an SEO strategy?
The unit of success is different. An SEO strategy optimises for a position in a list a person clicks. An AI SEO framework optimises for being the source a model quotes, which can happen with no click at all.
On our own site over 90 days, 6,442 page-one queries produced zero clicks between them. Judged as rankings that looks like failure. Judged as citation it’s the channel working.