Last month I cancelled my entire Semrush subscription, worth around $3,600 a year, and replaced the part I actually needed with a tracker I built myself using the DataForSEO API. This was not a rage quit. It was the result of a known bug, a broken promise, and the realisation that AI search visibility is now too important to outsource to a tool that cannot measure it correctly. Here is what happened, and how you can build your own tracker for a fraction of the cost.
Why I cancelled a $3,600 Semrush subscription
I have been a Semrush customer and advocate for years. I have spoken at more than 10 webinars organised by Semrush and recommended the platform to thousands of marketers. I was not just a paying customer. I was the kind of customer their influencer budget is designed to create.
Then their Prompt Tracking feature broke, and the way they handled it changed my mind about the whole platform.
The bug: brand mentions reported as missed
Prompt Tracking is Semrush’s tool for monitoring whether your brand appears in AI search results across ChatGPT, Perplexity, and other assistants. The core job of the product is identifying brand mentions.
Except it does not recognise common brand name variations. Legitimate mentions get reported as “missed”, which means the data is incomplete and every metric built on top of it is skewed. If a tool tells you that AI assistants are not citing your brand when they actually are, that is worse than having no data at all. You will make strategy decisions based on a false negative.
I reported it. Support confirmed, in writing, that the issue is on their end: my campaign was configured correctly, the fault sits in their own brand extraction logic, their development team is investigating, and there is no timeline for a fix.
The retention failure
Semrush initially applied a partial credit, acknowledging the problem. Then, with the bug still open and no fix date (Aug 5th 2026), they resumed charging the full $298 per month and refused a refund or continued discounted billing. As the same defect remains unresolved, I do not think it is reasonable to charge the full subscription price again.
So I cancelled everything and disputed the charge with my credit card provider.
The irony is hard to miss. Semrush spends heavily on influencer marketing, paying well-known SEO experts to promote the platform, while a long-term advocate walked away over an acknowledged bug and a promise that was not kept. Customer acquisition is expensive. Retention is usually the cheaper growth strategy, and it fails quietly, one support ticket at a time.
A year ago this would have frustrated me for weeks. Today it is just another API integration.
What an AI search tracker actually needs to do
An AI search tracker is a monitoring system that records whether AI assistants such as ChatGPT, Perplexity and Gemini mention or cite your brand when answering user prompts, which sources they select, and how that changes over time.
Before building anything, it is worth being precise about the job. Tracking visibility in AI search is a different problem from tracking rankings, because LLMs do not rank pages, they select sources. A useful tracker needs to answer four questions:
- Does the assistant mention my brand when users ask the prompts my customers actually ask?
- Which sources does it cite when answering those prompts, and am I one of them?
- How does this change over time, so I can connect content work to visibility gains?
- Where do competitors appear that I do not?
The Semrush bug broke question one, which breaks everything downstream. Brand mention detection is the foundation of the entire measurement stack. If your tool cannot match “Victoria Olsina”, “victoriaolsina.com” and “Victoria Olsina Growth Marketing” to the same entity, your visibility score is fiction.
How I built my own tracker with DataForSEO
DataForSEO is an API-first data provider. There is no dashboard subscription and no seat pricing. You pay per request, which for a monitoring workload is dramatically cheaper than $99 per brand per month.
The core components
Prompt set. Start with the 20 to 50 prompts that matter commercially: the questions your ideal clients ask an assistant before they would ever find your site. For my practice that includes prompts about Web3 SEO consultants, GEO for crypto brands, and AI search visibility services.
AI search results. DataForSEO’s AI optimisation endpoints return structured responses from LLM platforms for your prompts, including the text of the answer and the cited sources. Each run costs cents, not a monthly fee.
Brand extraction you control. This is the part Semrush got wrong, and the part you can get right in an afternoon. Write your own matching logic: exact brand name, domain, common misspellings, name variations, and founder names. Because you define the entity variants yourself, nothing gets reported as missed when it was actually there.
Storage and trends. Log every run to a spreadsheet or a small database with the date, prompt, platform, mention status, and cited sources. After a few weeks you have the trend line Semrush was charging $298 a month to draw incorrectly.
Scheduling. An automation platform such as Make or n8n runs the whole loop weekly without anyone touching it. This is the same systems-first approach I use for AI content and marketing automation: build the workflow once, let it run daily or weekly, review the output monthly.
Semrush vs DataForSEO for AI search tracking
| Criteria | Semrush Prompt Tracking | DIY tracker with DataForSEO |
|---|---|---|
| Pricing model | $99 per brand, per month | Pay per API request, cents per prompt check |
| Monthly cost, 50 prompts weekly | $99+ | Single-digit dollars |
| Brand mention detection | Automated extraction with a known, unresolved variant-matching bug | Matching logic you define, covering every brand variant |
| Data ownership | Locked in their dashboard | Your spreadsheet or database, exportable forever |
| Setup effort | None | An afternoon, plus a Make or n8n automation |
| Best for | Teams wanting zero setup and accepting current data gaps | Teams that need accurate mention data and cost control |
What it costs
My tracker costs single-digit dollars per month in API calls. Semrush wanted $99 per brand per month for the same job, done wrong. Even including the build time, the payback period was under one billing cycle.
The bigger lesson: own your measurement layer
The point of this story is not that Semrush is a bad company. Their keyword and backlink data remain solid, and for many teams the all-in-one platform is still the right choice.
The point is that AI search tracking is a young category. The incumbents are bolting it onto rank-tracking architecture built for a different problem, and the cracks show. When the measurement layer for your most strategically important channel is unreliable, owning it beats renting it.
This matters more if you are a Web3 or crypto brand, because AI assistants are already answering your buyers’ questions whether you are cited or not. Before you invest in tracking, check whether AI systems can even read your site properly with the free AI visibility checker. Tracking mentions is pointless if your site is technically invisible to the crawlers feeding the models.
Conclusion
I did not plan to build a tracking product. I planned to keep paying a vendor to do it. But when a tool’s core function is broken, the vendor admits it, and the bill arrives anyway, building your own stops being a technical flex and starts being the rational choice. If your AI search data comes from a black box, ask your vendor the question I asked mine: how exactly do you match brand variants? The answer will tell you whether your visibility numbers are real.
Frequently Asked Questions
What is an AI search tracker?
An AI search tracker monitors whether AI assistants such as ChatGPT, Perplexity and Gemini mention or cite your brand when answering user prompts. Unlike a rank tracker, it measures source selection and brand mentions rather than positions on a results page.
Is DataForSEO cheaper than Semrush for AI search tracking?
For monitoring workloads, yes, by a wide margin. DataForSEO charges per API request, typically cents per prompt check, while Semrush’s Prompt Tracking costs $99 per brand per month. A weekly tracking loop across 50 prompts costs single-digit dollars monthly.
Can I build an AI search tracker without being a developer?
Mostly, yes. Automation platforms such as Make or n8n can call the DataForSEO API, run brand matching, and log results to a spreadsheet without custom code. The only genuinely technical decision is defining your brand variant list, which is marketing knowledge, not engineering.
Why do AI search tools miss brand mentions?
Most rely on automated brand extraction logic that fails on name variations, domains used as brand references, and partial matches. When the extraction layer misses a variant, the mention is logged as absent, which skews every downstream visibility metric.
Want your brand cited by AI search engines, not just ranked by Google? My LLM SEO for Web3 service makes crypto and Web3 brands discoverable and citable inside AI-generated answers. Book a call to find out where your brand stands today.
