Search "apa itu reksa dana" on Google right now. An AI Overview appears, explains mutual funds in four paragraphs, and cites three sources.
Now search "harga saham BBRI hari ini".
No AI Overview. Just results.
Same category. Same regulator. Same searcher, possibly the same session. And Google made a deliberate decision to summarise one and step back from the other.
Understanding why is worth more than any keyword list.
The Pattern, Not the Percentage
You will read that AI Overviews appear on some specific percentage of financial queries. Somewhere between eight and ninety-one percent, depending which article you happen to open.
All of those numbers are real, and none of them mean what they appear to mean.
They measure different keyword sets. A tracker sampling "finance keywords" broadly gets one figure. A tracker sampling educational finance queries specifically gets another. The category-wide percentage is an artefact of the sample, not a property of the category.
What holds across every measurement is the shape.
AI Overview trigger rate, by financial query type
BrightEdge Generative Parser. Google summarises where synthesis helps, and withdraws where a stale answer costs someone money.
That is the line. Not a percentage. A judgement, made by Google, about where an AI-generated summary might cause harm.
Why Google Steps Back From Stock Prices
Consider what an AI Overview actually is. A synthesis of multiple sources, generated at query time, presented as an answer.
For "apa itu reksa dana", that synthesis is straightforwardly useful. Mutual funds worked the same way last week. They will work the same way next week. Nothing about summarising the concept can hurt anyone.
For "harga saham BBRI hari ini", synthesis is dangerous. The correct answer changes every second the market is open. An AI Overview generated from cached sources could show a price from twenty minutes ago, and someone might act on it.
This is Google's own Your Money or Your Life caution, visible in a product decision rather than in a guidelines document. Financial content sits in the strictest quality tier precisely because inaccuracy damages financial stability. When Google cannot guarantee accuracy, it declines to synthesise.
Local intent behaves similarly. "Bank terdekat" wants a map, not a paragraph. Summarising it adds nothing and risks sending someone to a branch that closed.
The Uncomfortable Implication
Educational content built the top of every financial marketing funnel in Indonesia.
Explainers about mutual funds. Guides to mortgage calculation. Articles defining inflation, compound interest, credit scores. That content attracted the audience, established authority, and fed people into product pages.
It is now the content most likely to be summarised without a click.
Where the exposure sits
Most exposed
Educational explainers. Definitional content. Concept guides. The top of the funnel, summarised at 91%.
Partly exposed
Rate explanations. Calculation guides. Planning content. Summarised roughly two thirds of the time.
Barely exposed
Real-time data. Local branch queries. Anything Google cannot safely summarise from cached sources.
The content that attracted your audience is the content Google is now most willing to answer on your behalf.
Nobody planned for this. The educational content strategy was correct for a decade, and the ground moved under it.
What Survives Summarisation
The instinct is to abandon educational content. That instinct is wrong, and the reason is worth sitting with.
An AI Overview cites its sources. Being cited inside one is worth more than ranking beneath one, because the citation carries the brand into the answer itself.
So the question is not whether to publish educational content. It is what kind of educational content an engine cannot compress away.
Original data
An engine can summarise the definition of a credit score. It cannot summarise your institution's own analysis of approval rates by income bracket, because that analysis exists nowhere else and cannot be synthesised from other sources.
Publish something only you know, and the summary has to cite you or omit the fact entirely.
Regulatory citation
The first peer-reviewed study of Generative Engine Optimization, published at ACM KDD 2024, tested ten thousand queries across nine domains. Adding source citations, quotations and statistics improved visibility by 30 to 40 percent. Keyword stuffing produced no effect.
The domain that benefited most from statistics was Law and Government, the closest proxy for regulated financial content.
Quoting POJK 22/2023 directly, with the article number, is content an engine wants to cite because it is verifiable. And it is content most competitors do not produce, because paraphrasing feels easier than quoting.
Calculation methodology
Publishing a number is weak. Publishing how the number was derived is strong, because the methodology is quotable and the number alone is not.
A page that says "effective interest is roughly 22%" invites a summary. A page that shows the calculation, explains why flat and effective rates diverge, and cites the transparency requirement that mandates disclosing both, gives an engine something it must attribute.
Where the Content Should Move
If the top of the funnel is being answered without a click, funnel value has to migrate downward.
Migrating funnel value
Comparison content
Product against product, with the tradeoffs stated honestly. Harder to summarise, and closer to a decision.
Eligibility transparency
Who qualifies, who does not, and why applications get rejected. Nobody publishes this. Everybody searches it.
Interactive calculators
A summary cannot replace a tool. Build them in-page rather than inside an iframe, where nothing gets indexed.
Proprietary analysis
Data only your institution holds. The engine must cite you or leave the fact out.
None of this abandons the educational layer. It stops relying on it for traffic that no longer arrives.
The Volatility Nobody Mentions
One more thing, and it changes how often you should look.
BrightEdge tracks citation change across categories. Financial services is the most volatile they measure. More than half of financial domains see their citation position change week to week, and roughly nine in ten of those changes are declines.
Week to week. Not quarter to quarter.
An annual AI visibility audit in this category measures a position that stopped existing in week three. That is not a service upsell. It is arithmetic.
September 2025 demonstrated what happens when a retrieval weight changes without announcement. Reddit's citation share inside ChatGPT collapsed. Five different trackers measured the collapse and reported five different figures, from 14.29% down to 0.21% by one measure, from roughly 60% down to 10% by another, because they measured different quantities. All five were correct. The channel lost most of its value in six weeks, and nobody was told.
Financial content lives in the category where that kind of shift happens most often.
What This Means for the Two Disciplines
AI Overviews sit at the boundary where traditional search and generative engines stop being separable.
The content that earns a citation inside an AI Overview is the content that satisfies Google's quality raters: named authors, verifiable credentials, regulator citations, transparent methodology. The same signals, serving both purposes.
Which means fintech SEO and Generative Engine Optimization for financial services are not competing budget lines. The technical foundation must exist before an engine can cite anything, and the authority signals that rank a page are the ones that make it citable.
I wrote Found Before They Search partly because institutions kept asking which to invest in, and the question contained a false premise. The companion volume, Cited or Silent, covers how engines decide what to cite in categories where being wrong causes harm. Both are free as gated editions, and both are published in paperback and hardcover and listed on Google Play Books and Apple Books.
Where to Look First
Take your ten highest-traffic educational articles. Search each of their target queries.
Count how many now show an AI Overview. Count how many of those Overviews cite you.
The gap between those two numbers is what the next twelve months of work is about.
Frequently Asked Questions
What percentage of financial queries trigger AI Overviews?
The question has no single answer, and any figure quoted without its keyword set is misleading. BrightEdge data shows roughly 91% on educational finance queries, 67% on rate and planning queries, about 10% on local intent, and 7% on real-time stock price queries. Other trackers report different category-wide percentages because they sample different keyword sets. The pattern is the finding, not the percentage.
Why does Google withdraw AI Overviews from stock price queries?
Because synthesis from cached sources could show a price from twenty minutes ago, and someone might act on it. Financial content sits in Google's strictest quality tier precisely because inaccuracy damages financial stability. When Google cannot guarantee accuracy, it declines to summarise. The same logic applies to local branch queries, where a summary risks sending someone somewhere that closed.
Should we stop publishing educational content?
No. An AI Overview cites its sources, and being cited inside one carries your brand into the answer itself. The question is what kind of educational content an engine cannot compress away: original data only you hold, direct regulator citation with article numbers, and transparent calculation methodology that the engine must attribute rather than paraphrase.
What content is least exposed to summarisation?
Comparison content with honestly stated tradeoffs. Eligibility transparency explaining why applications get rejected, which nobody publishes and everybody searches. Interactive calculators, built in-page rather than inside an iframe where nothing gets indexed. Proprietary analysis of data only your institution holds.
How often should we monitor AI citation in financial services?
Weekly. BrightEdge tracks financial services as the most volatile citation category, with more than half of financial domains changing position week to week and roughly nine in ten of those changes being declines. An annual audit measures a position that expired in week three.
Can a citation source really disappear overnight?
September 2025 demonstrated it. Reddit's citation share inside ChatGPT collapsed after a retrieval weight change, with no announcement. Five trackers measured it and reported five different figures, from 14.29% falling to 0.21% by one measure, from roughly 60% to 10% by another, because they measured different quantities. All five were correct. The channel lost most of its value in six weeks.
Sources & References:
- BrightEdge Generative Parser - AI Overview trigger rates by financial query type, and week-on-week citation volatility across categories.
- Google Search Quality Rater Guidelines - YMYL classification for financial content.
- Aggarwal, Murahari, Rajpurohit, Kalyan, Narasimhan, Deshpande - "GEO: Generative Engine Optimization", ACM SIGKDD 2024, arXiv 2311.09735.
- Spotlight, SEMrush, PromptWatch and RBC Capital - four independent measurements of Reddit's September 2025 citation collapse inside ChatGPT, reporting different figures because each measures a different quantity.
- POJK 22/2023 - Otoritas Jasa Keuangan, consumer protection provisions.
- Arfadia Digital Indonesia - AI Citation Rate Report 2026. arfadia.com/resources