Google AI Overviews rewired the SERP (and your funnel)
Google AI Overviews changed what “ranking” even means because the first interaction is often an answer, not a list of ten blue links. If your page is not a source the model can confidently quote, you can sit in position three and still lose the click.
Two patterns show up repeatedly when we audit SaaS sites that saw a traffic dip in 2025-2026:
First, queries with “how”, “best”, “vs”, “pricing”, “templates”, and “examples” are increasingly satisfied directly in the results. That is not a theory, it is visible in Search Console as impressions holding steady while clicks drop. Second, the winners inside AI Overviews tend to be pages that read like reference material: clear definitions, scoped claims, and fast verification.
Google has been public about the direction for years: it wants content that is “helpful” and written for people, not search engines. The difference in 2026 is enforcement. AI Overviews made low-trust pages easier to ignore at scale. If you have not read it recently, Google’s own guidance on creating helpful, reliable, people-first content is basically a checklist for “will an AI system feel safe citing this?”
A practical reframe that helps teams: you are not only optimizing for a click. You are optimizing to become the cited source, then earning the click later through brand recall, comparison intent, and product-led follow-ups.
The death of thin pages: topical depth beats keyword density
Topical depth vs keyword density is not a slogan. It is how you survive when AI systems compress the web into summaries.
Thin pages fail in predictable ways:
They answer the query “enough” but not completely, so they are interchangeable. They repeat the keyword but do not add unique constraints, examples, or data. They do not define terms cleanly, so they are hard to quote. And they often lack a point of view, so there is no reason to pick them as a canonical source.
When we rebuild these pages, we usually do less publishing, not more. A content team might replace 30 “feature” posts with 8 pages that each own an entity and its surrounding questions: definitions, decision criteria, implementation steps, pitfalls, and benchmarks. That structure gives Google and LLMs something stable to model.
If you want a quick diagnostic, pull 20 pages that lost clicks and ask one question: “Does this page contain at least one standalone paragraph that would make sense if quoted out of context?” If not, it is not written for AI summaries, and it is probably not written for humans either.
If your current workflow produces generic drafts, fix the inputs before you blame the outputs. This is why we keep a library of constraints and tone rules, not just prompts. VellumUp’s breakdown of AI writing prompts that actually improve output is a good starting point for teams trying to get consistent, non-fluffy structure.
Why E-E-A-T signals now outweigh backlink volume
E-E-A-T signals are not a direct ranking factor you can toggle, but they map to the risk model Google is using to decide what to surface in AI-driven results. In 2026, backlinks still matter, but a high link count cannot rescue low credibility the way it used to.
Here is what we see working in real audits, especially for B2B SaaS:
Experience and expertise show up as specificity. Not “we help teams scale”, but “we migrated 14 SaaS blogs from subdomains to subfolders and recovered 38% of lost non-brand clicks in 6 weeks.” Those details are hard to hallucinate and easy to trust.
Author and site transparency matter more when the query has financial, health, or operational risk. Clear author bios, editorial policies, and “last updated” practices reduce ambiguity. Google’s rater guidelines still frame how quality is evaluated at scale, even if the raters do not directly change rankings. The concept is stable.
Original data and reproducible claims are the cheat code. A small benchmark table, a methodology note, or even a screenshot-based walkthrough can outperform a generic “guide” with ten times the word count.
This is also where entity authority shows up. If your brand is consistently associated with a topic across multiple pages, and those pages get referenced, discussed, and revisited, you build a stronger “aboutness” signal than a random batch of keyword targets.
If you are fighting robotic tone because multiple writers or tools are involved, solve that now. We wrote a tactical fix for teams dealing with voice drift: brand voice matching to fix robotic AI blog posts.
Structured content wins: how to get pulled into AI summaries
Structured Data helps, but “structured content” is bigger than schema. It is the physical layout of information so a system can extract the answer with high confidence.
Answer Engine Optimization starts with predictable formatting:
A one-sentence definition near the top. A short “when to use this” paragraph. A step sequence with constraints (time, cost, prerequisites). A comparison table. A pitfalls section. And a tight conclusion that states the decision rule.
That is how you become quotable.
The extraction-friendly page pattern we use (and why it works)
This is the pattern that most often gets cited in AI summaries in our tests across SaaS and ecommerce:
- Define the term in one sentence, no hype.
- Give a “best for” and “not for” boundary.
- Provide steps with concrete inputs and outputs.
- Add a small table that compares options or outcomes.
- Close with a verification checklist (what to measure in Search Console or analytics).
Order matters. Models and humans both prefer early clarity, then detail.
Schema is still worth implementing, especially for FAQs, how-tos, software apps, and organizations. Google’s reference on structured data and rich results is the canonical baseline. Just do not confuse “adding schema” with “being the best answer.” Schema amplifies a good page, it rarely redeems a weak one.
Here is a simple table we use to align writers and SEOs on what “structured enough to cite” looks like:
| Page element | What it does for AI Overviews | What it does for humans |
|---|
| One-sentence definition | Creates a safe quote | Confirms they are in the right place |
| Comparison table | Reduces ambiguity | Speeds up decisions |
| Step sequence with prerequisites | Enables procedural summarization | Prevents failed implementations |
| Pitfalls and edge cases | Signals real experience | Builds trust and reduces churn |
| Clear “next action” | Connects answer to intent | Improves conversion rate |
If you are publishing at scale, structure is where automation should help, not hurt. VellumUp’s website scan explanation of what AI learns from your URL shows the kind of site-level patterns that make structured publishing consistent.
LLM citation factors: what gets cited, what gets ignored
Generative Engine Optimization is about increasing the probability that an LLM chooses you as a source when it synthesizes an answer. You cannot control the model, but you can control how easy you are to trust and extract.
In practice, LLM citation factors tend to cluster around four signals:
Verifiability: Claims that can be checked quickly win. Numbers with context, named frameworks, explicit assumptions, and links to primary sources all help.
Uniqueness: If your page is a paraphrase of the top five results, the model has no reason to cite you. Original examples, templates, and hard-earned edge cases create differentiation.
Entity clarity: If it is obvious who you are, what you do, and what you are qualified to say, you get safer citations. This is where consistent terminology and internal linking matter.
Content authority signals: Not just backlinks, but mentions, repeat visits, branded queries, and consistent topical coverage. Think “share of model”: how often your brand shows up in the model’s learned representation of a topic.
A clean way to test this: pick ten queries you care about and run them in multiple AI search surfaces. Track which domains are cited repeatedly. You will notice they are not always the highest DR sites, they are the clearest and most referenceable.
For the “DR” confusion that comes up in SEO teams: DR is Ahrefs’ Domain Rating, a third-party metric. It can help you gauge link competition, but it is not Google’s scoring system. Ahrefs explains the math behind it in their Domain Rating (DR) definition, which is worth sharing internally so stakeholders stop treating it like a KPI.
AI-first indexing: what changes technically (and what doesn’t)
AI-First Indexing does not mean Google replaced crawling with a chatbot. It means the downstream systems that interpret, summarize, and rank content are more dependent on clean inputs.
Technically, the fundamentals still apply: crawlable pages, correct canonicals, fast performance, and sane internal linking. The change is that messy sites now pay a larger penalty because AI features prefer high-confidence extraction.
Three technical issues we see killing AI Overview visibility even when “SEO basics” look fine:
JavaScript-heavy rendering that hides the main answer until late in the DOM. Over-aggressive canonicalization that collapses distinct intents into one URL. And templated pages that look unique to humans but are near-duplicates to a classifier.
If you publish automatically, your indexing workflow has to be just as automated as your writing. Otherwise you end up with “published” content that never becomes eligible to rank. Even if you are not on Wix, the debugging logic in fixing indexing issues after auto publishing is the same: validate crawl, validate canonical, validate sitemap, then validate coverage in Search Console.
What optimizing for AI search means for your content calendar
Optimizing for AI search is not “write longer posts.” It is publishing fewer, stronger assets that can win citations and conversions.
This is the calendar shift we recommend to SaaS founders and lean content teams who cannot afford waste:
You build topic clusters around entities, not keywords. Each cluster has one flagship page that defines the entity and solves the core job-to-be-done, plus supporting pages that answer sub-questions with tight scope. Internal links are intentional, and every page has a unique role.
You prioritize traffic quality over quantity. A page that brings 400 visits but converts at 2.5% beats a page that brings 4,000 visits and converts at 0.1%, especially when AI Overviews reduce top-of-funnel clicks.
You plan content around “citation moments.” These are queries where the model needs a stable source: definitions, benchmarks, step-by-step implementations, and comparisons. If your content is built like reference material, AI systems will use it.
A simple planning table keeps teams honest:
| Content type | Primary goal in 2026 | Best use case |
|---|
| Definition + implementation guide | Earn citations and rank | “What is X and how do I do it?” |
| Comparison page | Capture decision intent | “X vs Y”, “best X for Y” |
| Benchmark or study | Build authority | “Average cost/time/impact of X” |
| Template or checklist | Drive saves and links | “X checklist”, “X template” |
If your current process is “brainstorm 50 keywords, assign writers, hope,” you will keep bleeding time. You need a system that researches, outlines, writes in your voice, and publishes on schedule without manual copy-paste. That is exactly what VellumUp automates via connected platforms and webhooks. If you are evaluating tools, start with how to choose the best AI for writing in 2026 so you do not buy something that only drafts text and leaves the hard parts to you.
Frequently Asked Questions
What does DR stand for in SEO?
DR stands for Domain Rating, a third-party metric from Ahrefs that estimates the strength of a site’s backlink profile. It is useful for competitive research, but it is not a Google metric and it does not guarantee rankings.
What is DR and PR in SEO?
DR is Ahrefs’ Domain Rating. PR usually refers to PageRank, Google’s original link-based algorithm concept; Google no longer exposes public PageRank scores, and modern ranking systems use many signals beyond links.
What is Dr. and DA in SEO?
DR (Ahrefs) and DA (Moz Domain Authority) are both third-party metrics that approximate link-based authority. Use them to size up competition, not as success metrics for your own site.
What is the 80/20 rule for blogging?
In practice, about 20% of your posts tend to drive 80% of results. In 2026, that usually means a small set of authoritative, structured pages that earn citations, links, and conversions while the rest contribute marginally.
Next step: rebuild one page for AI Overviews, then scale the pattern
Pick one high-intent page that already gets impressions in Search Console but has declining clicks. Rewrite it to be citation-ready: one-sentence definition, clear boundaries, step sequence, a comparison table, and a short pitfalls section. Once that page starts earning better rankings and mentions, scale the exact structure across a full entity-based cluster.
If you want to automate the whole pipeline (research to publishing) while keeping your brand voice consistent, start by connecting your site and running a scan. Create your account at VellumUp registration for automated SEO publishing and publish your first AI-optimized article from a single pasted URL.
VellumUp11 min read
