This is the tool stack I use to run SEO and AI visibility work for my clients, written for agency owners and freelancers deciding what an AI-native setup needs. The short version: 6 tools, with Claude in the middle and every data source wired into it through MCP or an API, so the model reads live Search Console, crawl and AI-answer data instead of exports.
I built it this way because of what client audits keep showing me. In August 2026 I ran a paid AEO audit for a Swiss health insurer. Perplexity cited 177 URLs across the audit's prompts, and only 14 of them were on the insurer's own site: 8%. 3 comparison portals held 26%. A rank tracker and a backlink tool would have reported a healthy website and missed most of where the answer came from. (The full audit, anonymised.)
AI-native means the model does the research and I make the calls
In my setup, Claude Code is connected to each client's Google Search Console, their analytics and AEO Copilot. Once a week it pulls what changed. I spend my own time on the pages closest to revenue, the ones that turn a buyer into a customer, and on the judgment calls a model gets wrong: positioning, priorities, what the client will actually ship.
In practice that changes 3 things:
- Audits run on demand. The model can re-query a site's data whenever I ask, so I don't wait for a quarterly export.
- AI answers get tracked next to rankings. Whether ChatGPT, Claude, Perplexity and Google AI Overviews name the client is its own metric, with its own prompts.
- Repeating checks run on a schedule. One of my routines checks every month whether my best AEO tools roundup has gone more than 35 days without an update, and emails me to re-verify every tool's price. I stopped relying on memory for it.
Rank trackers measure your client's pages, AI answers come from other people's
Most SEO tools were built for keywords and backlinks, and they still do that job well. They can't tell you what an AI engine says when a buyer asks for a recommendation, or which sites it built the answer from.
The insurer audit made this concrete. On process questions such as how to switch hospital insurance, ChatGPT named no insurer at all. It answered from comparis.ch, moneyland.ch, krankenkassenvergleich.ch, bonus.ch and the federal portal ch.ch. None of that appears in a rank tracker.
My own brand had the same blind spot. When I started tracking AEO Copilot on 27 August 2026, the baseline was 1 mention across 28 answers (7 prompts, 4 engines), and the page cited was my LinkedIn article, not the website. What that baseline taught me shaped the content I wrote next.
The 6 tools in my stack, and what each one feeds Claude
- Google Search Console, connected through an MCP server, so Claude queries real click, impression and indexing numbers instead of summarising an export.
- Claude as the reasoning engine: it reads the other 5 sources and drafts the audit.
- Keywords Everywhere for search volume and related queries, pulled straight into the conversation.
- PageSpeed Insights for Core Web Vitals, so technical recommendations rest on this week's numbers.
- Firecrawl (firecrawl.dev) to crawl the client site and its competitors and extract structured content without manual scraping.
- AEO Copilot for AI answer tracking: which prompts mention the client, in which engine, and which pages the engines cite.
With the Search Console MCP connected, I can ask Claude which pages lost clicks this month and it queries the property itself. No CSV, no copy-paste. The same goes for every tool on the list: each one turns a manual research step into data the model can read directly.
The skills that drive this are open source: 21 agent skills for SEO and AEO, including the baseline visibility report and the weekly pulse I run for clients.
Legacy vs AI-native agency, side by side
| Legacy SEO agency | AI-native SEO agency |
| Audit cycle | Days, manual, quarterly | Hours, on demand |
| Main visibility metric | Search rankings | Rankings plus AI answer mentions and citations |
| Content production | Human draft, human edit | Draft grounded in live data, human review |
| Analyst time | Mostly pulling and formatting data | Mostly strategy and judgment |
| Tooling | Rank trackers, backlink databases | Same signal sources, wired in through MCP, plus an AI answer tracker |
| Client reporting | Static monthly decks | Reports regenerated from live data |
My monthly client audit, step by step
- Pull real data first. The Search Console MCP server gives Claude live click, impression and indexing data.
- Crawl and diagnose. Firecrawl pulls the client site and its top competitors. Claude flags technical issues, content gaps and cannibalisation.
- Check AI answers. I start each new client with 5 non-branded prompts where they should clearly be mentioned, plus a separate set of branded prompts for reputation. On one client, a pharma brand, branded visibility was 100% while ChatGPT called the drug "not a recognized medication" (the full example).
- Prioritise. Claude turns the findings into a ranked list of fixes, weighted by effort and impact. Money pages go first.
- Human review. I cut anything that doesn't fit the client's positioning and rewrite what stays in the client's language.
The output is a report the client can read. I generated one for Webflow's AI visibility through this workflow, and Flowsultants laid it out as a 6-page client PDF: here is how that report was built.
Freelancers and small agencies gain the most
The research bench used to cap how many accounts a 1- or 2-person shop could take. With the data work handled by the model, the cap moves. Agency accounts on AEO Copilot track more than 10 client brands on average (account data, September 2026). Larger agencies gain consistency instead: the same audit, run the same way, across every account.
If you work solo, answer engine optimization is a service you can sell, and the audit above is a good first offer.
Where AEO Copilot fits
I built AEO Copilot, so weigh my take on it accordingly.
It covers the AI answer layer: whether a brand gets named and cited in ChatGPT, Claude, Perplexity, Gemini and Google AI Overviews. The free plan tracks 1 brand and 50 prompts on ChatGPT and Claude. The Startup plan at $9 a month covers 4 engines of your choice and adds the API and MCP access this stack depends on. The Agency plan is $99 a month for unlimited brands and 600 prompts, with no per-brand fees. For other tools and their dated prices, see Best AEO Tools in 2026.
FAQ: AI-native SEO agency tools
What makes an SEO agency AI-native?
Where AI sits in the workflow. In an AI-native agency, the model pulls and joins the data (Search Console, crawls, AI answers) and the strategist decides what to do with it. Using ChatGPT to draft blog posts on top of a manual process is a different thing.
What does an AI-native tool stack include?
Mine has 6 tools: Google Search Console through an MCP server, Claude as the reasoning engine, Keywords Everywhere for keyword data, PageSpeed Insights for performance, Firecrawl for crawling, and AEO Copilot for AI answer tracking.
How long does an AI-native audit take compared with a legacy one?
A legacy technical audit typically takes days of manual data pulling. With the data wired into Claude, the pulling part takes minutes and the audit fits in a few hours. Most of that time goes to review and prioritisation.
Do small agencies or large agencies benefit more?
Small ones feel it first. A 1- or 2-person team loses the research bottleneck that limited how many clients it could take. Large agencies use it to keep audits consistent across accounts.
What does AEO Copilot add that SEO tools don't?
Tracking of AI answers: whether ChatGPT, Claude, Perplexity and Google AI Overviews mention a brand for a given prompt, how they describe it, and which pages they cite. In my insurer audit, 92% of the URLs Perplexity cited were not on the client's site, which is the part SEO tools can't see.
To see the gap on a real client, run a free AI visibility audit and compare it with what their rank tracker says.