What GEO Gets Wrong About Content

The most popular GEO tactics you will find on agency blogs, including 'mention your brand three times per page,' 'add an FAQ to every page,' and 'rewrite all your headers as questions,' are not supported by any published research from Google, Bing, or the original academic paper that started this whole conversation. The honest foundation of generative engine optimization is ranking well in organic search first. Everything else is secondary, and most of it is overrated.

The Short Answer Most GEO Guides Skip

Most GEO guides are built on a misreading of one paper, extrapolated into a checklist by people who have a service to sell. The tactics that actually move citations in AI-generated answers are the same fundamentals that have moved organic rankings for years: topical depth, verifiable data, and genuine authority signals. If you want the practical craft of getting cited in AI answers, I have covered the foundational mechanics in detail elsewhere. This article is specifically about separating what research actually supports from what agencies invented and dressed up as best practice.

The frame here is simple: I am going to tell you what Aggarwal et al. (2023) actually measured, what it actually found, and where the agency checklist version diverges from reality. Then I am going to give you a working checklist of what to do instead, including the tradeoffs nobody in this space wants to acknowledge.

What the Actual Research Says (and What It Does Not)

The source that most GEO content cites is the 2023 paper 'GEO: Generative Engine Optimization' from researchers at Princeton and Georgia Tech (Aggarwal et al., 2023). It is a real paper with real findings, and those findings are being consistently misrepresented.

What the paper actually measured: citation visibility lift in AI-generated answers when specific content modifications were applied. The modifications that produced statistically significant results were adding verifiable statistics (up to 40% increase in citation visibility) and including authoritative citations within the content itself (up to 30% increase). Those are the findings.

What the paper did NOT validate: brand repetition, FAQ stuffing, keyword density manipulation, or conversational header rewrites as standalone tactics. None of those showed statistically significant lift in the paper's methodology. The agency checklist version of GEO took the genuine findings about statistics density and authoritative sourcing, buried them, and led with the tactics that are easier to sell as a recurring service.

The practical implication is blunt: if your GEO strategy centers on brand mentions and FAQ schema, you are doing work that has no published evidence base. If it centers on citing primary sources like the CFPB, Freddie Mac's Primary Mortgage Market Survey, or HMDA data within substantive content, you are working from something real.

The Ranking Prerequisite Nobody Wants to Hear

About 99.5% of pages cited in Google AI Overviews already appear in the top 10 organic results for the same query, based on BrightEdge's 2024 analysis of over 100,000 AI Overview citations. That single number should end most GEO conversations before they start.

If your target page ranks in position 15, no amount of schema markup, FAQ sections, or conversational headers is going to put it in an AI Overview. The AI is pulling from what Google already trusts. You cannot shortcut your way around that with content-layer tactics.

For mortgage content specifically, this gap is structural. AI Overviews and ChatGPT citations in the lending vertical skew heavily toward the CFPB, HUD, FHA, Bankrate, NerdWallet, and Investopedia. Independent mortgage brokers and regional lenders are almost entirely absent from AI-generated answers, even when their content is accurate and well-written. This is not a keyword density problem. It is a domain authority and referring domain problem that takes years to close.

The answer for regional lenders is not to compete head-to-head with national aggregators on 'what is a 30-year fixed mortgage.' It is to build fast, crawlable pages that AI systems can actually parse on locally specific and product-specific queries where Bankrate has thin or generic coverage. That is a realistic ceiling, and it is worth knowing before you commission a GEO audit.

What Actually Moves the Needle: A Practical Checklist

Here is what I actually recommend, split by whether you need developer access.

Developer Access Required

Structured data as a parsing aid, not a citation trigger. FAQ schema, HowTo schema, and Article schema help Googlebot parse semantic meaning more reliably. That indirectly supports inclusion in AI training data crawls and real-time retrieval. But schema does not cause a citation. If the underlying content is weak, schema on top of it does nothing.
Entity relationship density. Pages that clearly define and connect related concepts in a single coherent context perform better in AI retrieval than pages that repeat a primary keyword without building the conceptual map around it. For an ARM loan page, that means defining the ARM loan, the rate cap structure, the initial adjustment period, and the lifetime cap in one place, with clear logical relationships between them. This is an architectural decision, not a copywriting one.
Canonical tags and crawl hygiene. AI retrieval systems inherit authority from Google's index. If your site has duplicate content, orphaned pages, or canonical tag errors, you are bleeding trust signals that directly affect AI citation probability.
Page speed. AI crawlers behave similarly to Googlebot in preferring fast, efficiently rendered pages. A JavaScript-heavy page builder template adds real overhead here.

Content Layer Only (No Developer Required)

Cite at least one primary source per page. CFPB, Freddie Mac PMMS, HMDA data, MBA weekly application data. Freddie Mac's Primary Mortgage Market Survey appears in AI-generated mortgage answers far more frequently than any lender-produced rate content. Synthesizing and citing that data is the single highest-ROI content change available without touching code.
Place a direct answer within the first 150 words of any section that poses a question. AI retrieval is extractive. It looks for a question pattern followed immediately by a concise, standalone answer. A FAQ buried at the bottom of 3,000 words is far less effective than a structured answer at the top of the section.
Add author credentials, a cited sources list, and a last-updated date. Google's Search Quality Evaluator Guidelines are explicit that E-E-A-T signals matter more in YMYL verticals like mortgage and lending. These are content-layer changes with zero technical barrier.
Build entity relationships in copy, not just in schema. Write content that defines related concepts and connects them explicitly. 'An ARM loan's rate cap limits how much your interest rate can increase at each adjustment period and over the life of the loan' is more useful to an AI retrieval system than a page that says 'adjustable rate mortgage' forty times.

The Compliance Tradeoff Generic GEO Guides Ignore

Here is where most GEO advice for the mortgage vertical falls apart: it is written by people who have never had to run copy past a compliance officer.

CFPB disclosure requirements, Regulation Z rate advertising rules, and state licensing constraints are not optional. They create real limits on what you can publish, how you can phrase it, and what data you can display without triggering disclosure obligations. The advice to 'add a rate table to your pages to earn AI citations from rate-comparison queries' is technically naive. Displaying rates on a mortgage website without the required APR disclosures, loan assumptions, and regulatory boilerplate is a compliance violation in most states.

The practical framework I use: treat GEO best practices and compliance requirements as two separate constraint sets. Where they agree, the choice is easy. Where they conflict, compliance wins, always, without exception. The GEO tactic that creates regulatory exposure is not worth the citation.

For mortgage marketers, this means the highest-value GEO work concentrates on educational content, data synthesis from primary sources, and entity-rich explanatory pages, not on rate display or product-specific advertising copy. That is a narrower runway than generic GEO advice suggests, but it is the one that will not land you in front of a regulator.

Where This Goes Wrong: Honest Tradeoffs

Even the evidence-based tactics have failure modes, and I want to name them plainly.

The measurement problem is real. Google Search Console now includes AI performance reports, which helps. But distinguishing between 'GEO work caused this citation increase' and 'my rankings improved and citations followed' is genuinely difficult. If you implement primary source citations and entity-dense copy at the same time you fix a technical SEO problem, you cannot cleanly attribute the citation lift. Track changes in a simple spreadsheet with dates, and give any single change at least 90 days before drawing conclusions.

There is also a quality gap between citation types. Appearing as a source link in a Perplexity results panel is different from being the quoted passage in a ChatGPT answer. The second is far more valuable and far harder to achieve for a regional lender. Most GEO reporting conflates the two.

And bloated page builder output that adds rendering overhead will undercut all of this. If your site is built on a heavy page builder that produces slow, JavaScript-dependent pages, the content-layer improvements I am recommending here have a lower ceiling. The infrastructure matters.

Timeline expectations: meaningful AI citation improvement for a regional lender working from a clean technical foundation takes six to twelve months minimum. Anyone promising faster results is selling something.

A Diagnostic You Can Run Without a Developer

If you are a mortgage marketer without developer access, here is a practical audit you can run on your existing pages today.

Step 1: Pull your top 20 pages by organic traffic in Google Search Console. Filter for pages ranking in positions 1 through 10. These are the only pages worth optimizing for AI citations right now. Everything else needs ranking work first.

Step 2: For each qualifying page, check three things:

Does the page cite at least one primary source (CFPB, Freddie Mac PMMS, HMDA, MBA)? If no, add one.
Does each major section that poses a question have a direct answer in the first 150 words of that section? If no, rewrite the section opening.
Does the page include author credentials, a last-updated date, and a cited sources list? If no, add them.

Step 3: Prioritize by gap size. The page that fails all three checks gets worked on first. A page that already cites primary sources but lacks a direct answer placement is a 30-minute fix. Start there.

Decision rule: if a page ranks below position 10, skip it entirely until the ranking problem is solved. GEO optimization on a page that ranks on page 2 or lower produces no measurable AI citation lift. That is the boring answer, and it is correct.

The one thing to do right now: open Search Console, filter to pages ranking in positions 1 through 5, and check whether any of them cite a primary government or institutional data source. That audit takes 20 minutes and will tell you exactly where your highest-ROI GEO work is.