When Google Can Rank Without an Index, SEO Stops Being “Position” and Becomes “In or Out”
DeepMind’s latest ranking research points to a future where search engines don’t retrieve lists—they generate the small set of results they’ll consider. If the “beam” is the universe, page two disappears, tracking rankings breaks, and SEO becomes the discipline of earning consistent inclusion across AI-driven answers.
Search has always had an invisible layer most businesses ignore: what the system considers before it ever “ranks.” A new line of research suggests that invisible layer may become the entire game.
Google DeepMind researchers have now shown—at least in theory and limited experiments—that a single language model can produce rankings without relying on a separate, classical search index. If that direction ships (even partially), the practical consequence for businesses isn’t a small algorithm update. It’s a redefinition of visibility: less about “position #3” and more about whether your brand is included at all in the small set an AI system generates as it answers.
This editorial is my attempt, as Marius Dosinescu at AYSA.ai, to translate that research into a business reality: what changes, why it matters, what can go wrong, and what operators should do now—without waiting for a crisis or a new acronym.
Table of contents
- The concise summary (what changed, why it matters, what to do)
- Context: how search got here (and why this isn’t “just another paper”)
- What “ranking without an index” actually means (in plain English)
- Documents become codes: why distinctness starts to matter more than ever
- The “beam” idea: why page two disappears
- How this changes customer behavior (even if your industry doesn’t care about SEO)
- Measurement when rankings become binary
- A concrete SME scenario: a local clinic and the new inclusion problem
- Content strategy in an inclusion-first world (what to stop, start, and keep)
- Technical SEO still matters—just in a less obvious way
- What agencies should rethink (deliverables, reporting, and pricing)
- Where AYSA fits: monitored, approved, executed—at the speed AI search demands
- What to do next (operator’s checklist)
- Sources and further reading
The concise summary (what changed, why it matters, what to do)
What changed: Research covered by Search Engine Journal suggests a future where a single generative model can produce rankings without relying on a traditional separate index, and where the ranked list becomes an internal computation rather than a user-visible “page of results.” Source: Search Engine Journal.
Why it matters: If a generative system only produces a small set of candidates (think top 5–10, not top 100), then “page two” is not a UI choice—it simply doesn’t exist. Visibility becomes closer to binary inclusion (“in the generated set” vs. “not generated”), and a lot of today’s Rank tracking, “Keyword positions,” and incremental on-page tuning becomes less representative of what customers actually experience.
What to do now (without overreacting):
- Shift reporting from position to inclusion: start building baselines for where your brand is cited, mentioned, or recommended in AI-driven search experiences.
- Optimize for distinctness: reduce semantic duplication across pages and product/service variants; make each page uniquely “about” something.
- Operationalize fast fixes: in an AI-first world, the winners are the teams who can notice visibility shifts early and ship corrections quickly—without breaking the site.
At AYSA.ai, our bias is execution: we monitor, prepare changes, ask for approval, and then implement accepted improvements—because strategy without shipping becomes invisible in the new search reality.
Context: how search got here (and why this isn’t “just another paper”)
To understand why this matters, you need to zoom out. Search used to be “retrieval”: crawl the web, store it in an index, and retrieve matching documents. Ranking was a scoring layer on top.
Over the last decade, the stack has been steadily moving from lexical matching (“does this page contain these words?”) toward semantic understanding (“is this page about what the user means?”). That shift has already changed outcomes for businesses:
- Exact-match keywords matter less than topical coverage and intent match.
- Thin pages that exist purely to capture variants (near-duplicate location/service pages) tend to collapse in value.
- Brand and entity understanding matters: Google needs to know who you are, what you do, and why you’re credible.
Now add generative AI to the surface area. For users, the experience becomes less “choose from a list” and more “get an answer now.” For businesses, that raises one brutal question: if the user doesn’t browse, how do you get chosen?
The SEJ piece frames this in a way that should make every operator uneasy: if ranking becomes something a model generates internally—and only generates a small set—then visibility is no longer “move from #8 to #4.” It becomes “exist in the generated set or don’t exist.”
This isn’t a claim that Google has replaced its index tomorrow. It’s a directional signal: research investment tends to follow product constraints, especially cost and latency. When research points to cost reductions and architectural simplification, it’s worth paying attention.
What “ranking without an index” actually means (in plain English)
In today’s simplified mental model, search works like this:
- Find candidates fast: retrieve a large list of possible pages quickly (the “cheap” step).
- Re-rank carefully: take a smaller set and rank it more accurately with more expensive computation (the “smart” step).
- Show a results page: users scroll, compare, click.
The research discussed in the SEJ article suggests collapsing parts of this into a single generative model that can output ranked identifiers directly (and in some experiments, rank items provided “in context”). That’s technical—so here’s the practical translation:
- Instead of retrieving a huge list and sorting it, the model generates a short list.
- If it doesn’t generate you, you’re not “ranked #37.” You’re absent.
- The output space becomes constrained by what the model can generate efficiently.
This aligns with a larger UX trend: when the interface is an answer card, a conversational response, or an AI overview, the notion of “100 results” becomes an internal artifact, not a customer experience.
And if you think, “Fine, but there will always be a list underneath,” remember: product surfaces follow economics. If the system can compute fewer candidates at lower cost and still satisfy users, the incentive to display long lists gets weaker.
Documents become codes: why distinctness starts to matter more than ever
One of the most important implications in the SEJ analysis is that documents can be represented as short identifiers (docIDs) that the model generates token-by-token—similar to how it generates words. That’s a conceptual shift:
- In classic SEO, you optimize a URL/page to rank for queries.
- In a generative ranking architecture, you’re competing to be the “address” the model chooses to generate.
Why that changes everything: a generative system needs pages to be separable in meaning. If two pages are near-duplicates—same intent, same claims, same structure—then they’re not two “separate competitors.” They are two nearly identical options occupying the same semantic territory.
For SMEs, near-duplicate patterns are extremely common:
- Service-area pages: “Plumber in Austin,” “Plumber in Round Rock,” “Plumber in Cedar Park” with 90% the same copy.
- Ecommerce variants: multiple product pages where only the color changes, with copy pasted from the manufacturer spec sheet.
- Agency landing pages: “SEO for dentists,” “SEO for lawyers,” “SEO for HVAC” that differ only in the heading.
Historically, you could get away with this because Indexing and ranking systems could still list many similar pages, and users might click around. In an inclusion-first world, semantic duplication is a direct visibility risk: you don’t just underperform—you get excluded.
Business takeaway: Your website needs fewer “keyword pages” and more meaningfully distinct pages—each earning the right to exist with unique information, evidence, and intent fit.
The “beam” idea: why page two disappears
Here’s the concept that should reframe how you think about search visibility: generative systems often use a decoding method that keeps only a limited number of candidate continuations at each step. Practically, that means there is a built-in cap on how many options are considered and returned.
If the system only ever produces a top-k set (say 5, 10, or 20), then:
- There is no “rank 51.”
- There is no page two.
- Below-the-fold optimization matters less than “make the cut.”
This matters because so much of SEO culture is built around the idea that you can climb positions gradually. If the system generates a short set, then visibility is discontinuous: you’re either included or you’re not.
What this does to planning:
- Incremental improvements still matter, but the goal shifts from “move up 2 spots” to “cross the inclusion threshold consistently.”
- Being “pretty good” across many pages can be worse than being the single best page for a topic cluster.
This is one reason I’m skeptical of strategies that optimize only for aggregate traffic while ignoring brand/Entity Coverage and AI citations. You can’t budget your way into inclusion if the site’s information architecture and distinctness are broken.
How this changes customer behavior (even if your industry doesn’t care about SEO)
Most business owners don’t care how search works. They care whether the phone rings, whether bookings happen, and whether product demand shows up.
In an AI-first experience, the default behavior changes in predictable ways:
1) Less browsing, fewer comparisons
If users get an answer that feels complete, they will click less. Not because they hate your website—because they don’t need it.
That doesn’t mean your website becomes irrelevant. It means your website becomes upstream infrastructure: it must be credible, machine-readable, and distinct enough to be selected as a source, citation, recommendation, or provider.
2) “Winner takes most” becomes more literal
In classic search, being #6 could still generate business. In a world where the interface shows 3–5 cited sources and a couple of actions (call, book, buy), the distribution gets steeper. That creates pressure in categories where margins are tight.
3) Trust becomes a UX problem, not a ranking factor
When users see fewer options, they have fewer ways to sanity-check. This increases the importance of:
- clear policies (returns, refunds, cancellations),
- credentials and licensing where relevant,
- reputation signals (reviews, awards, third-party mentions),
- accurate business details (location, hours, service area).
Not because those are “ranking factors” you can game—but because they help systems and humans decide you’re safe to recommend.
Measurement when rankings become binary
For 20 years, SEO reporting has leaned on a fragile proxy: keyword rankings. Rankings were never perfect, but they were legible. If you moved from #9 to #4, you could usually connect that to traffic and revenue.
If the future is “in the generated set or not,” then rank reports can become misleading in three ways:
1) The rank you track may not be what users see
Users may see an AI overview, a short list of cited sources, and maybe a reduced set of links. Your “#3” might appear nowhere if the AI answer is satisfied without it.
2) Below-the-beam ranks aren’t ranks
If only a limited set is generated, “rank 20” may be an artifact of an SEO tool’s simulation, not an actual impression opportunity.
3) You need inclusion baselines, not positional deltas
What should you measure instead?
- Inclusion rate: for a set of query themes, how often are you cited/linked/mentioned in AI-driven results?
- Page referenced: which URL or entity is used as the supporting source?
- Query class coverage: are you visible only on “brand” queries, or also on problem/solution queries?
- Consistency over time: do you appear reliably, or does visibility drift week to week?
This is exactly why we built AYSA capabilities around AI search visibility as a first-class concept—not an afterthought to rank tracking.
Important caution: I’m not claiming any single platform can “measure all AI results perfectly.” The ecosystem is fragmented and changing. The correct approach is to pick a representative set of customer questions and monitor presence, sources, and patterns over time—then ship improvements.
A concrete SME scenario: a local clinic and the new inclusion problem
Let’s make this real with a business that doesn’t have time for SEO theory: a multi-location physical therapy clinic.
Before: classic SEO mindset
- Create a page for each location: “Physical therapy in North Dallas,” “Physical therapy in Plano,” etc.
- Add service pages: “Back pain,” “Knee pain,” “Sports rehab.”
- Track rankings for: “physical therapy near me,” “PT Plano,” “back pain therapist Dallas.”
- Celebrate incremental ranking lifts.
After: inclusion-first reality
A user asks an AI-driven search experience: “Do I need physical therapy for knee pain after running? What should I do first?” The system produces an answer, plus 3–5 sources, plus a “book an appointment” style action set.
The clinic’s problems are no longer “we’re ranking #7.” They’re:
- The AI cites national health sites, not local providers. The clinic is absent.
- All location pages are nearly identical. The system can’t tell which one is most relevant or authoritative.
- Provider credentials exist, but are buried. The machine-readable footprint of expertise is weak.
- Appointment, insurance, and process info is unclear. Users can’t take action confidently.
What the clinic should do (practically)
- Build distinct, helpful pages: not 30 near-duplicates, but fewer, stronger pages that answer real questions with clinician-reviewed content.
- Make expertise legible: clear clinician bios, credentials, and review policies; link them properly across relevant pages.
- Create action-ready content: “What to expect,” “When to see a PT,” “Insurance accepted,” “Typical timelines,” “Red flags,” etc.
- Monitor inclusion: track a set of patient questions weekly and see whether the clinic is cited, and if so, which page is referenced.
This is not “SEO theater.” It’s customer acquisition and trust infrastructure—optimized for how discovery is evolving.
Content strategy in an inclusion-first world (what to stop, start, and keep)
If your content strategy is built on publishing volume and hoping Google sorts it out, an inclusion-first environment will punish you. The goal is not to publish more. The goal is to become the best candidate for the limited set that gets generated.
Stop: content that exists only to catch keyword permutations
- near-duplicate city/service pages with swapped headings,
- “SEO landing pages” that provide no unique evidence,
- manufacturer copy pasted across product pages,
- thin blog posts that reword what everyone else already wrote.
Start: content designed for selection
Think like a system that must decide what to include in a short set. It will favor sources that are:
- specific: clear topic focus, fewer mixed intents per page,
- original: unique data, unique examples, unique experience,
- verifiable: citations, policies, credentials, transparent claims,
- structured: easy for machines to parse and summarize,
- maintained: updated when reality changes (pricing, inventory, hours, regulations).
Keep: the fundamentals that prove you’re real
Even if “ranking factors” become less legible, the basics remain essential:
- clean information architecture,
- fast, accessible pages,
- clear Internal linking,
- credible off-site signals (PR, reviews, partnerships),
- accurate business data everywhere it appears.
There is no shortcut here. Inclusion is earned through clarity and trust, not just SEO mechanics.
Technical SEO still matters—just in a less obvious way
When people hear “AI Search,” they often assume Technical SEO becomes irrelevant. That’s backwards. Technical SEO becomes the baseline for eligibility and correct interpretation.
Here’s what I’d prioritize for SMEs and lean teams:
1) Crawlability and index hygiene (still foundational)
Even if ranking architectures evolve, systems still need access to your content. Broken internal links, blocked resources, and inconsistent canonicals are silent inclusion killers.
2) Structured data where it improves understanding
Use structured data thoughtfully to clarify:
- organization and identity,
- products, pricing ranges (when accurate), availability (when maintained),
- locations, hours, services,
- authors and credentials (where relevant).
Don’t treat structured data as a cheat code. Treat it as a schema for clarity.
3) Performance and UX
If AI surfaces reduce clicks, the clicks you do get become more valuable. Your site needs to convert quickly: speed, clarity, and action paths (call, book, buy).
4) Change control becomes a competitive advantage
In a volatile environment, you’ll need to fix and iterate without breaking production. That’s why we emphasize a monitored, approved execution loop rather than random one-off changes.
AYSA is built around this operational need: monitoring, change proposals, approval, and implementation—so improvements actually ship.
What agencies should rethink (deliverables, reporting, and pricing)
If you run an agency, this shift is both threat and opportunity.
Deliverables: from “pages shipped” to “coverage achieved”
Old deliverables:
- X blog posts per month
- Y backlinks per month
- Z keywords tracked
New deliverables (better aligned with inclusion):
- Query theme maps tied to customer journeys
- Inclusion/citation monitoring for representative prompts
- Content consolidation plans to eliminate semantic duplication
- Entity clarity improvements (bios, policies, trust pages)
- Conversion path upgrades for the clicks you do get
Reporting: stop anchoring clients to average position
Rankings are familiar, but familiarity is not accuracy. Clients will demand clarity in the face of volatility. That means building a reporting stack that includes:
- inclusion and citation patterns,
- query class performance (brand vs non-brand, informational vs transactional),
- impact on leads/sales, not just “visibility.”
Pricing: execution speed becomes billable value
When the environment changes faster, the ability to implement improvements safely—without dev bottlenecks—becomes a real differentiator.
This is where AYSA’s AI SEO tools can function as an execution layer for agencies: monitor, propose, approve, deploy—consistently—without turning every optimization into a Jira fight.
Where AYSA fits: monitored, approved, executed—at the speed AI search demands
The temptation in moments like this is to chase the newest idea and rebuild everything. That’s not how most businesses win.
Businesses win by doing the fundamentals better and faster than peers:
- keeping content accurate and distinct,
- maintaining technical hygiene,
- monitoring real visibility where customers search,
- shipping improvements consistently.
That’s the gap AYSA is designed to close.
The AYSA loop (how execution should work now)
- Monitor: track site changes, visibility signals, and AI search inclusion patterns over time. See AI search visibility and monitoring.
- Prepare: generate prioritized change proposals: consolidation opportunities, clarity fixes, internal linking, schema improvements, content upgrades.
- Approve: you control what ships. No black-box auto-publishing.
- Execute: implement accepted website changes quickly and safely—so the strategy becomes real.
This “approved execution” model matters more as search becomes more dynamic and less transparent. When you can’t rely on stable rankings, you rely on operational excellence.
If you want to explore how this fits your business size and risk tolerance, start with pricing, then review our latest thinking on the AYSA blog.
What to do next (operator’s checklist)
This is the part most editorials skip. Here’s an action list you can hand to a marketing lead, agency partner, or yourself.
In the next 30 days
- Define 30–50 customer questions across your funnel (problem, comparison, “best,” pricing, brand, local intent).
- Baseline AI inclusion: for those questions, record whether your brand/site is cited/linked/mentioned and which pages show up.
- Audit duplication: identify pages that are meaningfully the same (location variants, product variants, service variants).
- Fix top trust gaps: add or improve: About, Contact, policies, credentials, location details, and any “proof” pages that make you safer to recommend.
In the next quarter
- Consolidate and strengthen: merge weak/duplicate pages into fewer, better resources that are truly distinct.
- Create “selection-ready” assets: original FAQs, comparison pages, process explainers, inventory/pricing clarity (where feasible), and expert-reviewed guides.
- Improve internal linking: make it obvious which page is authoritative for each theme.
- Build an execution cadence: weekly monitoring, biweekly changes, monthly review.
Ongoing
- Watch for drift: inclusion can change without your site changing. Track patterns.
- Operationalize updates: new products, new services, changed hours, updated policies—publish and propagate consistently.
- Don’t outsource accountability: agencies and tools are great, but your business is the source of truth.
Sources and further reading
- Search Engine Journal: Google Now Has The Math To Rank Without An Index, And The Results Page Does Not Survive It (primary source for this editorial’s research input)
- Search Engine Journal: AI Search coverage
- Search Engine Journal: SEO coverage
- Search Engine Journal: Technical SEO coverage
- Search Engine Journal: Paid Media coverage
Note: The SEJ article references multiple research threads and implications. In this editorial, I’ve treated forward-looking claims as analysis, not certainty. If you want this to be operational—not speculative—start by building inclusion baselines and fixing distinctness issues. Those moves help regardless of which architecture wins.
If you’re building for AI-era discovery, not 2016-era rank chasing, begin with these AYSA resources:
Continue the AI search topic inside AYSA.
Use these pages to connect the article with AI SEO tools, AI visibility monitoring, AI Overviews and approved website execution.
Turn this topic into a website action plan.
Use these AYSA hubs to move from reading to technical fixes, AI visibility monitoring, research, glossary context and approval-first SEO execution.
