Same variations, same answer
For years, our keyword research strategy rested on a simple principle: the more query variations you cover, the more traffic you capture. Each different phrasing was treated as a distinct opportunity. "Best marketing agency," "top marketing agency," "good digital marketing agency" were three opportunities, three potential pages, three positions to track.
The tools we all relied on reinforced this worldview. Ahrefs, Semrush, Google Keyword Planner: they all reported separate volumes for each variation. We built spreadsheets with hundreds of keywords, grouped by intent, and planned content to cover them systematically. More keywords covered means more traffic captured. Simple. Elegant. And, as it turns out, incomplete.
AI models changed the equation. Not because they eliminated keywords, but because they treat variations as rephrasings of the same question. They compress dozens of formulations into one answer. And in that answer, they recommend nearly the same brands every time.
We've discussed the shift from ranking to citation before. But recent data puts the problem in sharper focus: the rules didn't just change. The playing field was always smaller than we thought.
What the numbers say: 91.6% convergence
An audit published by Search Engine Journal in August 2026 analyzed 3,750 responses from three different AI models across 250 commercial category queries. The results are telling.
The three models agreed on the top brand in just 41.6% of cases. Sounds low, but look at the next number: in 91.6% of cases, at least two of three models named the same top brand. The models don't agree on exact ordering. They agree on who's eligible. The set of recommended brands is small and stable. Only the order shifts around.
Even more relevant: a test of 2,000 runs across ten different buyer personas showed that category leaders maintained roughly 80% consistency regardless of persona. Mid-market brands, by contrast, swapped recommendations in up to 75% of cases depending on the persona.
In plain terms: if you're a category leader, AI recommends you no matter who it's talking to. If you're in the middle of the pack, the recommendation depends on context, and that's a much more fragile position.
Convergence didn't shrink the map. It revealed it.
Here's the insight that changes the conversation for the clients we manage at difrnt.: the space of real commercial opportunities was always smaller than the space of phrasings. Convergence didn't compress it. It made it visible.
Think of classic keyword research as a low-resolution map. You saw many dots, each query variation looking like a separate spot on the map. AI models cranked up the resolution and showed that many of those dots were actually the same place, viewed from different angles. No opportunities were lost. It just became clear how many there truly were.
Another experiment cited in the study confirms the mechanism. Researchers from Trine University and Texas A&M created product sets with one real brand and nine fictional brands, all with identical reviews, prices, and descriptions. The real brand was recommended in all 670 valid trials across three models, two languages, and four categories. Not a single fictional brand surfaced.
AI models don't read pages and classify them. They recognize entities and describe them. If your brand doesn't exist as a recognized entity, no amount of page optimization compensates. You could have impeccable content across 50 keyword variations, but if the brand isn't present as a distinct entity in the models' training data, you simply don't exist in their answers.
Only 8% of categories still have open space
The audit examined how many categories present a genuine "competitive vacuum": category queries where no brand dominates. The answer: just 8% of the 250 queries. And they weren't evenly distributed. The highest vacuum rate (20%) appeared in healthcare technology, a sector where the reference brand hasn't consolidated yet.
For digital marketing, the categories are largely settled. If you work in e-commerce, SaaS, or fintech, chances are AI models already have a fixed set of brands they recommend for category queries. And if you're not among them, you're not losing a battle. You're not even in it.
There's a nuance worth noting for emerging markets. In Romania, where we operate from, convergence is less advanced than in English-language markets. AI models have less high-quality Romanian content to build answers from, which means competitive vacuums are more frequent in Romanian-language queries. That's a real window of opportunity, but it's closing as models train on more local content.
What to do differently starting Monday
At difrnt., we've started applying a few concrete shifts in our clients' SEO strategies:
Check whether you're fighting for settled phrases. If three AI models already give the same answer to your main query and that answer doesn't include you, the cost of contesting that phrase is higher than you think. You're not just wasting budget on that phrase. You're losing the phrase you didn't contest instead. We tested this with a SaaS client: the primary phrase they'd been targeting for two years was completely settled across three international competitors. We redirected their content budget to a subcategory where convergence hadn't set in, and gained AI visibility within two months.
Invest in entity, not pages. The unit of investment in 2026 is no longer the optimized page. It's the brand name. Third-party citations, mentions in trade publications, presence on high-profile directories: these build the entity that AI models recognize. A robust Crunchbase profile, consistent press mentions, authored contributions on industry sites: these count for more than yet another optimized landing page.
Look for categories where convergence hasn't settled. Long, specific queries remain less stabilized. So do niche categories in emerging industries. This is where classic keyword research strategy still works, but the window is closing. Test with a simple check: ask the same query in ChatGPT, Claude, and Gemini. If the answers vary significantly across models, the category is still open.
Stop contesting phrases you've already lost. It sounds harsh, but it's pragmatic. Content budget allocated to phrases where AI has already settled on an answer doesn't produce incremental visibility. Redirect it toward categories where convergence hasn't locked in yet. This isn't about giving up on SEO. It's about doing SEO where it still has impact.
