Claude can audit your website, pull your Google Search Console data, and draft content. What it cannot do is see what ChatGPT, Perplexity, Gemini, or Grok actually answer when a buyer asks about your category, because Claude has no access to any of those AI search engines. A viral DIY workflow circulating on social media promises otherwise: a guide that turns a claude.ai project into a free AI search visibility system, with a site audit, AI citation checks, fix generation, a tracking dashboard, and a daily logging routine. It asks Claude to "test" those engines anyway, which means its citation results are guesses presented as data. This article breaks down what the workflow actually does, where each step quietly fails, what it gets right, and what a working version of the same loop requires.
The workflow matters because it is the new face of the most common AEO strategy: do it yourself with free tools. AI SEO guides like this one spread fast precisely because the promise, full AI search visibility for $0, is what every founder wants to hear.
The Bottom Line
- The DIY workflow's free setup cannot see AI search engines. Nothing in it connects to ChatGPT, Perplexity, Gemini, or Grok, so its citation checks produce simulated answers, not observed ones. Paid engine-response data exists, but daily checks across engines and queries get expensive fast.
- The automation is you. The workflow's tracking dashboard is a claude.ai artifact that refreshes only when you open it and approve the data pulls, its daily check-in is a prompt with no scheduler behind it, and the only live data wired in is your Google rankings. Every step runs when you run it.
- The generated fixes do not move AI visibility. The workflow's main deliverables are llms.txt files and schema markup. Google's AI optimization guidance dismissed llms.txt outright in May 2026, and the same guidance says structured data is not required for AI search.
- The content is written blind. Claude drafts with no knowledge base of your product, no competitive data, and no targeting of the queries where you are missing. The output is generic AI content of the kind AI search engines already deprioritize.
What the Viral Workflow Promises
The DIY workflow sets up a claude.ai project with a four-step loop: audit your site's technical SEO and Google rankings, check whether AI search engines cite you, diagnose keyword and content gaps, and generate fixes you deploy to your site. It connects Google Search Console through an open-source MCP integration, optionally adds a paid data API for Google rank checks, and finishes with a dashboard artifact plus a daily logging routine. Total setup is around two hours, and the pitch is that the whole system then runs itself.
The structure of that loop is sound. Audit, diagnose, fix, verify is roughly how any serious AEO platform works, including ours. The problem is not the shape of the loop. It is that three of the four steps have nothing real flowing through them.
Claude Cannot See What AI Search Engines Answer on Its Own
Claude has no access to ChatGPT, Perplexity, Gemini, Grok, or Google AI Overviews. When the DIY workflow prompts Claude to "test" whether your site shows up in those engines, Claude cannot run those tests. It can search the web through its own search tool, which is not any of those engines, or it can produce a plausible-sounding "cited: yes/no" verdict from general knowledge. Either way, you are reading a simulation formatted as a report.
The deeper failure is the diagnosis. The workflow asks Claude to state, for each engine, "the single reason why" you are not cited. Claude never observed the engine's answer, never saw which sources it pulled from, and never saw which sub-queries it ran. A reason invented after the fact leads to a prescription aimed at the wrong problem. AI search results also change between identical asks, and our research found that only 38% of citations persist from one week to the next, so even a genuine manual spot-check is a coin-flip sample, not a baseline.
The workflow's own fallback concedes the point: it suggests running your domain through free third-party visibility checkers when you want real data. Those checkers are one-shot snapshots from someone else's tool, which is the thing the workflow was supposed to replace.
There is one technically honest path to real data here. Data providers like DataForSEO sell AI search engine response data alongside their Google rank checks, and Claude can reach them over MCP. Wire that in and the citation checks become real. But the data is metered per response, and the workflow's daily cadence multiplies it: every engine, every query, every day is a separate billable call. By the time the checks are real and continuous, you are paying for a monitoring pipeline and operating it by hand, which is the product category the workflow claimed to replace. The free version simulates; the real version is neither free nor automated.
What to do: Treat any AI visibility claim that did not come from a live engine response as unverified. If you want to know whether ChatGPT recommends you, something has to actually ask ChatGPT, repeatedly, because single snapshots measure noise.
The Automation Does Not Exist
The DIY workflow promises a system that runs itself, and it cannot deliver one. Its "live dashboard" is a claude.ai artifact, and as of June 2026, artifacts on paid plans can connect to MCP servers, so the dashboard can genuinely pull Search Console data. But it does so only when you open it and approve the tool calls, nothing refreshes unattended, and the only live data wired in is Google rankings. The dashboard's AI visibility panels have nothing real behind them unless you add the paid engine-response data covered above.
The daily logging step is an instruction to schedule a prompt every morning at 8am. A claude.ai web project has no scheduler, so in the workflow as written, nothing fires at 8am. Claude's desktop app does offer scheduled tasks, and Claude Code can run cloud routines, but moving the prompt there does not fix the underlying problem: a scheduled run on the free setup is still logging Google ranking data plus simulated verdicts. The honest description of the system is a set of prompts you run by hand, with results you copy into a tracking file yourself. The workflow's own time estimate, 15 minutes per week of ongoing work, quietly admits this.
What to do: Be skeptical of any "autopilot" claim attached to a chat interface. Continuous monitoring requires infrastructure that runs whether or not you log in: scheduled checks, stored history, and alerting. If the system stops working when you stop prompting, it is a workflow, not an autopilot.
The DIY Fixes Are the Ones Google Debunked
The DIY workflow's fix-generation step produces three main deliverables: an llms.txt file, JSON-LD schema markup, and robots.txt rules for AI crawlers. As of June 2026, Google's own AI optimization guidance states you do not need to create machine-readable files like llms.txt, do not need to chunk or rewrite content specifically for AI, and that structured data is not required for generative AI search. No major AI search engine uses llms.txt for citation decisions, and schema markup does not affect whether ChatGPT cites you either.
This is the quiet irony of the workflow: its monitoring step produces fake data, and its fixing step produces real files that do nothing. A user can run the full loop, deploy everything, and have changed almost nothing about their AI search visibility.
What to do: Spend the deployment effort on what the evidence supports: content that directly answers the questions buyers ask AI search engines, published or refreshed within the last 30 days, on your site and on the third-party sources engines actually cite.
The DIY Content Is Written Blind
When the DIY workflow generates content, Claude is writing from whatever happens to be in the chat context. There is no knowledge base of your product's actual features, pricing, and positioning, so Claude can misstate your own facts inside the content you deploy. There is no competitive data, so the content does not know which competitor claims it needs to displace or what AI search engines currently say when buyers ask about your category. And there is no gap targeting, so the content is aimed at topics rather than at the specific queries where you are missing.
The result is generic AI content, and AI search engines are already deprioritizing exactly that. Volume does not rescue it: publishing more content does not fix AI visibility when the content is not grounded in what engines are actually answering. Content that wins recommendations is specific, references real competitive context, and answers a real intent that engines currently answer with someone else.
What to do: Before generating anything, establish two inputs: a factual source of truth about your own product, and evidence of what AI search engines currently answer for your target queries. Content written without both is guesswork with good formatting.
Your Own Website Is the Surface AI Search Engines Cite Least
Every fix the DIY workflow produces lands on your own website. But in our initial March 2026 citation research, only 6.3% of 1,122 citation URLs pointed to the tracked brands' own websites. The large majority of AI citations come from third-party sources: Reddit threads, YouTube videos, review sites, and editorial coverage. The workflow has no Reddit step, no YouTube step, and no way to discover which third-party sources each engine pulls from for your queries.
Per-engine differences make this worse. Our research found that Grok drives 60% of all Reddit citations across the engines we track, and when Grok went offline for a week, Reddit citations collapsed 91%. Meanwhile, as of May 2026, ChatGPT links to brand websites in 18 to 25% of its citations versus 8 to 9% for Grok. A workflow that only touches your own domain is invisible to the engines that lean on third-party sources, and a single generic "AI readiness score" cannot express any of that.
What to do: Find out which sources AI search engines cite for your category queries, then build presence there. For most categories that means Reddit threads engines already pull from and comparison content third parties have not written about you yet.
What the DIY Workflow Gets Right
Two ideas in the DIY workflow are genuinely good. Connecting Google Search Console and surfacing striking-distance keywords, queries where you rank in positions 8 to 20, is smart prioritization, because AI search engines retrieve from pages that already rank, and those queries are your fastest path from indexed to cited. And the loop itself, audit, diagnose, fix, track, is the right mental model for AEO. The author's instincts about the process are sound. The gap is that every step needs real data flowing through it, and the free version has none.
If you have the time and a content team, the DIY pieces worth keeping are GSC hygiene, striking-distance targeting, and fresh content aimed at buyer questions. That is real work that compounds.
What a Working Version of the DIY Loop Requires
The honest checklist for the audit, diagnose, fix, track loop the DIY workflow describes: live responses from each AI search engine on a schedule, the source URLs behind each answer, competitor names and claims extracted from real responses, content generated against a factual knowledge base and aimed at measured gaps, distribution across your site plus Reddit and YouTube, and rechecks after publishing to see whether visibility moved. Each piece exists to fix a specific failure covered above, which is why skipping any of them recreates the corresponding blind spot.
That checklist is what an AEO platform is. Loudmink runs this exact loop with real engine data: it checks AI search engines on a continuous cadence, shows which sources each engine pulled from, extracts competitor claims, generates reviewed content, and keeps monitoring after publication. Plans start at $99/mo as of June 2026 for ChatGPT tracking; five-engine coverage (ChatGPT, Gemini, Perplexity, Claude, Grok), Reddit, and YouTube execution come with higher tiers. And if you like working in Claude Code, that preference is right: Loudmink ships an MCP server, so Claude Code can read your real engine data, add queries, and score articles from the terminal. Claude is a capable operator. It just needs eyes.
Frequently Asked Questions
Can Claude or Claude Code check if my brand shows up in ChatGPT?
Not on its own. Claude has no access to ChatGPT, Perplexity, Gemini, or Grok, so any "citation check" produced by prompting Claude alone is a simulation. Real checking requires querying each engine directly, either through paid engine-response data APIs you wire in yourself or through an AEO platform that does it continuously.
Is the DIY Claude AEO workflow worth doing at all?
Parts of it. Connecting Google Search Console and targeting striking-distance keywords (positions 8 to 20) is genuinely useful, and writing fresh content aimed at buyer questions always helps. The monitoring, dashboard, and automation claims do not hold up, and the llms.txt and schema fixes have been debunked by Google.
Does llms.txt help my site show up in AI search?
No. Google's AI optimization guidance confirmed llms.txt is not needed, and no major AI search engine uses it for citation decisions as of June 2026. Content that answers buyer questions, kept fresh within a 30-day window, is what drives retrieval.
What does DIY AEO actually cost?
Around two hours of setup plus weekly manual work, with no verifiable AI visibility data at the end. Making the data real means paying for metered engine-response APIs, where checking every engine for every query every day adds a billable call each time. The hidden cost either way is decision quality: prescriptions based on simulated engine results can aim your content effort at the wrong gaps for months.
How do I connect Claude Code to real AI search data?
Use a platform that ships an MCP server. Loudmink's MCP connector gives Claude Code access to live engine tracking data, competitor analysis, and content scoring with one install command, so Claude works from observed answers instead of guesses.