The new SEO stack: what replaces your old tools
Summarize this article with AI
I’m going to tell you something agencies hate to hear
69% of AI Overviews cite your competitors. Not you.
Search Engine Land published the analysis in June 2026. I observe the same pattern every week in audits. A site with 800 product pages, 3,200 organic sessions per month. Meanwhile, AI Search recommends three competitors. Never once the brand.
Old tools don’t catch that.
Your Screaming Frog tells you your title tags are fine. Your Ahrefs shows a solid Domain Rating. Your Semrush alerts you to ranking shifts. But none of them tell you why Google mentions a competitor in the AI Overview. Because none read the Knowledge Graph in real time. Because none cross your entities with the ones the AI judges relevant.
The breakdown is in the stack.
I’ve been building systems since 2016. 1,300 semantic clusters later, I see a common thread: those who swap their old crawlers for scripts and LLMs unlock results dashboards never predicted. No magic. Just better-read signals.
The new SEO stack is a three-pillar architecture, not another layer on top of your SaaS subscription. I’ll show you here: a real client case, raw numbers, and the first script you can launch in 10 minutes.
What if your stack was built in 2018?
Your current stack doesn’t see AI Search
A B2B client calls me in March 2025. Catalog of 1,200 technical spec sheets. 18 months of « SEO content marketing. » One full-time writer. Premium crawler subscription. Result: 47 pages indexed out of 900 useful ones. 3,200 organic sessions per month. Zero rankings on strategic entities.
The crawler ran every week. Alerts on 404 errors. Duplicate reports. Zero analysis of entity coverage. The client thought they had a content problem. Actually, their semantic architecture was silent. Keywords weren’t the issue. Logical links between entities? They didn’t exist.
The old stack rests on four pillars:
- A crawler (Screaming Frog, Sitebulb)
- A rank tracker (Semrush, Ahrefs)
- A link tool (Majestic, Moz)
- A CMS that spits out pages with no internal linking logic
These tools analyze historical metrics: links, keywords, pages. They don’t capture entity structure or the conversational intent of LLMs. When Google rolls out AI Overviews, they keep measuring the classic SERP. And you’re staring at a wall.
Worse: these tools create dashboard fatigue. You spend Monday mornings parsing ranking fluctuations. Meanwhile, your competitors structure clusters, deploy validation scripts, test prompts that find gaps in the Knowledge Graph.
One number: among my clients who switched stacks, time spent on factual competitive analysis (not rankings, but entities cited in AI Overviews) dropped from 4 hours per week to 11 minutes. Because a script does the work.
The real problem is the blind spot in your tools, not the tools themselves.
LLMs: the brain your arsenal was missing
When I show a client a prompt that lists entities Google associates with their market, silence lasts three seconds. Then the question comes: « Why doesn’t my tool do that? »
LLMs don’t replace your expertise. They amplify your reading.
Take GPT-4o API or Claude 3.5 Sonnet. You feed it: content from your 200 most important pages, organic keywords extracted from Search Console, the list of competitors appearing in Google AI Overviews. You ask: « What entities does my site not cover, even though they appear in AI Google responses for my topic? » In 90 seconds, you get a list of missing entities with relevance scores. No traditional tool does this.
At a SaaS client, the LLM identified 23 « blind » entities in 4 minutes. The SEO team spent two content sprints covering them. Four months later, 14 of those entities triggered mentions in AI Overviews. And 6 generated optimized snippets. No new backlinks.
LLMs also validate your semantic clusters. In the DOSE framework I use (taught by Guillaume Attias at BMO Academy), the discovery and optimization step now relies on a prompt-result-validation cycle. I feed the LLM the planned cluster skeleton, it crosses entities against the public Knowledge Graph, and returns a coverage matrix. What used to take 7 hours of manual work is done in 22 minutes.
The trap to avoid: letting the LLM write your content without structure. Generated content works if structure is sound. Your architecture drives the result. The LLM is a composer, not an architect.
Concrete result: a software publisher saw organic traffic rise from 4,100 to 12,700 monthly sessions after using an entity prompt to rethink their sitemap. 82% pages indexed after 4 months. Versus 41% before.
Here’s how the new stack operates in practice, based on the case of an electrical component manufacturer. Each step replaces a blind spot of legacy tools.
The new SEO stack workflow
From entity extraction to continuous monitoring
APIs and scripts: your SEO nervous system
The second pillar of the new stack is APIs. Google Search Console, Ahrefs, OpenAI, Cloudflare, Google Trends. Connected via Python or Google Apps Script.
Why? Because SaaS tool interfaces filter raw data. You see what the tool wants you to see. With an API, you see signals you decide to cross.
An example: I built a 130-line script for an online learning platform. It:
- extracts Search Console queries weekly for the past 3 months,
- identifiés those whose click-through rate drops over 15% while ranking stays stable,
- queries the GPT API to check if an AI Overview appeared on the query,
- feeds a Google Sheet the client reads in 4 minutes Monday morning.
Before: 3 hours of manual research. After: 4 minutes. With 97% AI Overview detection reliability (tested on 200 queries).
APIs aren’t just for developers. I know SEO consultants using Make (formerly Integromat) or n8n to chain these calls with no code. The goal is to escape « monthly dashboard » logic.
Another script at an e-commerce auto parts seller crosses product XML feeds with the Semrush API to detect traffic-orphan pages. The script produces a CSV prioritized by cluster potential. 800 pages optimized in 6 weeks. Without this hack, it would’ve needed an agency and a 15,000 euro budget.
The script doesn’t erase the craft. It frees time for architecture. That’s the new stack: humans make decisions, machines do the grunt work.
1,200 spec sheets, zero structure: the numbered client case
March 2025. An electrical component manufacturer shows me their site. 1,200 technical spec sheets, a 40-article blog, an institutional homepage. Organic traffic: 3,200 sessions per month. Their goal: capture RFQs via highly specific queries.
Diagnosis: 47 pages indexed. 47 out of 1,200. The rest weren’t even crawled properly. The old agency talked about « cannibalization. » My read was simpler: zero architecture linking entities. Every spec sheet was an island.
I deployed a stack in three weeks:
- An entity extraction script via GPT API, crossed with electrical patents and standards in the domain.
- A semantic cluster in 4 levels (family, type, product, use case).
- An LLM to generate page meta-structures (not text, but Hn hierarchy, FAQs, entity lists to cover).
- A Python script tracking indexation via Search Console, with daily alerts on newly indexed or de-indexed pages.
Existing content was restructured. No massive rewrites. 6 weeks of work.
Result in 7 months: 12,700 organic sessions (that’s +297% from the starting 3,200; I round to +300% for messaging, but the raw figure is 297%). Better: 228 pages indexed. And 6 inbound RFQs via SEO, one of which closed for 47,000 euros last month.
The old stack would never have produced this. It would’ve kept churning ranking reports on keywords the site didn’t master at the entity level.
The new stack doesn’t work miracles. It makes visible what was already relevant.
Your first script in 10 minutes
No need to be an engineer. I built this chain for a client in 2025. It’s reproducible with a spreadsheet and an OpenAI API key.
Goal: identify missing entities across your 20 main pages.
Steps:
- Extract your 20 most important URLs from Search Console (Pages tab, last 3 months).
- For each URL, collect the text content (simple copy-paste into a CSV file).
- In Google Sheets, use the
=IMPORTDATA()function or Apps Script that calls the OpenAI API with this prompt: « List the 10 main entities on this page. Then, for each entity, indicate whether it’s likely to be recognized by Google’s Knowledge Graph. Reply in CSV format. » - Cross the returned entities with an export of AI Overviews observed in your sector (a basic Python script can query Google Programmable Search, or use a tool like AlsoAsked for questions).
Setup time: 10 minutes if you already use Google Sheets and have an API key. Result: an entity coverage matrix your competitors don’t have.
It’s an overlay. It reads what your tools ignore. And it changes everything.
One more number: at a software publisher that industrialized this script, covered entities rose from 34 to 117 in 5 months. AI Overviews noticed.
What if your stack was the obstacle?
I’ve seen too many sites vanish from AI results due to badly read signals, not failing content. The traditional stack measures the past. The new one reads the present.
My stack rests on LLMs, APIs, and scripts. None replace strategy. But they give diagnostic speed your competitors don’t have.
I see it every week from my desk in Southeast Asia. Clients who adopt this stack don’t spend more time on SEO. They spend it in the right place: architecture.
A client told me recently: « Before, I fixed errors. Now, I build bridges. »
Do you want to know if your pages are read by AI Search or just your crawler? Do you want to map your entities in 22 minutes instead of 3 days?
Then when do you change your stack?
Live audit of your semantic architecture in 45 minutes
I show you, page by page, the entities Google associates with your market and the ones you miss. My audit call is your action plan, live. No slides.
Book a strategic call — 45 minFrequently Asked Questions
What really replaces a tool like Semrush or Ahrefs in this new stack?
Nothing replaces them fully: they stay useful for market research. The decision layer that used their dashboards is replaced by scripts and LLMs. These cross your Search Console data, your entities, and AI Overviews.
Do you need to know how to code to use SEO scripts?
No. With Make, n8n, or Google Sheets + Apps Script, you chain API calls with no code. An SEO consultant can build their first entity detection script in under an hour.
Won’t LLMs risk inventing entities or data?
That’s a risk if you don’t cross outputs with verified sources (Search Console, Google Knowledge Graph API, or internal data). In the new stack, LLMs make hypotheses. Scripts validate them. This hypothesis-validation pair is what brings precision.
How much does setting up such a stack cost?
API costs are low: pennies per GPT call. Google Search Console and Google Sheets are free. The main cost is learning time. I estimate one day for an SEO familiar with CSV exports. Benefits? Hours freed every week.
Does this approach work for a small brochure site too?
Yes. A small site has less data, so scripts run faster. A simple Google Sheet with an entity validation script is enough to identify why a page isn’t appearing in AI results. Investment scales with catalog size.
