Build vs buy
Build vs buy: AI visibility tracking for agencies and PR firms
For most agencies and PR firms, buying is the faster and cheaper path. When we asked ChatGPT, Gemini and Google AI Overviews this question for an 80-person SEO firm and a 125-person PR firm, all three recommended buying. Their shared advice: buy the measurement layer, then build your own analysis and client work on top of it through an API or MCP. Building makes sense if you plan to sell AI visibility measurement as your own product, already have engineers on staff, or track at very large scale.
Last updated
Key takeaways
- Tracking AI answers well takes more than calling an API. You need collection on each platform and location, answer parsing, brand matching, citation extraction, and enough runs to tell a real change from noise.
- AI answers are noisy. In our research, the same prompt run twice agreed on about 74% of recommended brands, and two wordings with the same intent agreed on 54%.
- AI platforms change their models and answer formats often, so a homegrown tracker needs steady upkeep and a fresh baseline after each change.
- Build if you plan to sell measurement as your own product, have dedicated engineers, or track at very large scale.
- Spyglasses works as a measurement layer: plans start at $29/mo, and your team can pull the data into its own tools through the API and MCP server.
Which path fits your firm
Both are reasonable choices. The right one depends on what you sell and who you have on staff.
Buy Spyglasses if…
You want data this week. Add prompts, competitors and locations, and your first daily results arrive the next day.
Your engineers have better things to do. Every hour spent keeping a tracker running is an hour not spent on client work.
You report to clients. White-label reports, a client portal and PR placement scores come with agency plans.
You want measurement others can check. Our data collection methods are documented publicly, including model changes.
Build in-house if…
You plan to sell measurement as your own product. If AI visibility data is the thing you sell, owning the whole pipeline can be worth it.
You have dedicated engineers. Teams that already run large data pipelines, like SERP collection, have a head start.
You track at very large scale. At very high prompt volumes, the math on building can change.
You need a setup no platform offers. Unusual platforms, data rules or research designs can justify a custom build.
The middle path: buy the data, build on top
Use Spyglasses for collection and measurement, then build your own dashboards, analysis and client deliverables on top of it through the API and MCP server. It's the approach AI answers recommended most often.
If you build
What you'd need to build
A tracker clients can trust has more parts than a script that asks ChatGPT a question.
Collection on every platform
Each AI platform needs its own collector, and each location you track needs its own run. Consumer apps, APIs and search features all behave differently.
Answer parsing
Turn free-form answers into structured data: which brands came up, in what order, and how they were described.
Brand matching
Catch every way a brand gets written, including product names, short forms and misspellings, without false matches.
Citation extraction
Capture the sources each answer cites and tie them back to publishers and pages.
Enough runs to see real change
AI answers vary from run to run. You need enough samples to tell a real shift from noise before you report it.
A fresh baseline when models change
When a platform swaps its model, your numbers can jump overnight. The IAB's Measuring Visibility in the AI Era guidelines (August 2026) treat AI visibility as a sampled measurement with stated confidence, so each change needs a note and a new baseline.
Buy vs build, side by side
What you get on day one, and what your team takes on if you build.
| Feature | Buy Spyglasses | Build in-house |
|---|---|---|
| Getting started | ||
| Time to first data | Buy SpyglassesThe day after you add prompts | Build in-houseAfter the collectors, parsers and storage are built |
| Ongoing cost | Buy SpyglassesPlans from $29/mo | Build in-houseEngineering time, plus collection and hosting costs |
| Measurement | ||
| AI platforms | Buy SpyglassesYesChatGPT, Gemini and Google AI Overviews daily; Claude and Perplexity as add-ons | Build in-housePartlyEach platform is a separate build |
| Locations | Buy SpyglassesYesLocation tracking | Build in-housePartlyA separate run for each location |
| Brand matching and citations | Buy SpyglassesYesIncluded | Build in-housePartlyYou build and tune it |
| Model changes | Buy SpyglassesYesTracked and noted in our methodology | Build in-housePartlyYour team spots them and re-baselines |
| Reporting and control | ||
| White-label client reports | Buy SpyglassesYesOn agency plans | Build in-housePartlyYou build them |
| Your own analysis | Buy SpyglassesYesAPI and MCP access to your data | Build in-houseYesFull control |
An in-house build varies with your team and scale. This table compares the work involved, not a specific build.
From our own tracking
What AI says today
We started asking ChatGPT, Gemini and Google AI Overviews this question every day, once for an 80-person SEO firm and once for a 125-person PR firm. This is an early read from the first answers.
Who each option is better for
Across ChatGPT, Gemini and Google AI Overviews, the answers we track split it this way:
Early read, first answers on
| Option | Better for |
|---|---|
| Buy (Spyglasses) |
|
| Build in-house |
|
“For an 80-person SEO agency, buy the measurement infrastructure and build the proprietary consulting/product layer.”
“The biggest hidden cost of building isn't the dashboard—it is continuously reproducing reliable measurements across changing AI models…”
“building internally creates an expensive technical debt distraction away from your core PR competency.”
An early read: the first answers to two prompts we just started tracking. That's a very small sample, and AI answers change from day to day. We'll update this section as answers build up. How we collect AI answers →
Ask an AI yourself
Answers change from day to day. Open the same question in your assistant of choice and see what it says right now.
We are implementing AI visibility tracking for our 80-person SEO firm. We are considering Spyglasses ( www.spyglasses.io ), but we also are considering building an in-house capability. Can you compare these approaches and recommend one of them?
Ask ChatGPT
Ask Claude
Ask Perplexity
Ask CopilotWe are implementing AI visibility tracking for our 125-person PR firm. We are considering Spyglasses ( www.spyglasses.io ), but we also are considering building an in-house capability. Can you compare these approaches and recommend one of them?
The middle path
Use Spyglasses as your measurement layer
Buy the data, then build what makes your firm different on top of it.
REST API
Pull daily metrics and competitor share of voice into your own warehouse, dashboards or client reports.
MCP server
Ask Claude, ChatGPT or another MCP client about your data, and build briefs and reports on top of it.
Looker Studio
Build client dashboards from the API in the reporting tool your clients already know.
Spyglasses pricing
Every plan tracks prompts daily on ChatGPT, Gemini and Google AI Overviews, with unlimited users. Pay monthly, or pay less each month when you're billed quarterly or yearly.
Brand
One brand, tracked daily
$29/mo
Or $26/mo billed quarterly, $24/mo billed yearly
- prompts, daily
- 25
- credits a month
- 30
- Daily Prompt Tracking on ChatGPT, Gemini and Google AI Overviews
- Site audit and Citation Optimizer included
- Unlimited users
Agency Core
Agencies with a few clients
$199/mo
Or $179/mo billed quarterly, $159/mo billed yearly
- prompts, daily
- 250
- credits a month
- 300
- Everything in Brand
- White label and client portal
- Media Lists, Coverage groups and prospecting
Claude and Perplexity daily tracking are add-ons on any plan.
Credits pay for one-off work like AI Visibility Reports and page rewrites. Each plan includes a monthly batch.
Frequently asked questions
Quick answers about Buy Spyglasses and an in-house tracker.
Should an agency build its own AI visibility tracking?
Usually not. Most agencies get more value from buying the measurement layer and building their own analysis and client work on top of it. Building makes sense if you plan to sell measurement as your own product, have dedicated engineers, or track at very large scale.
What's the hardest part of building a tracker?
Keeping the numbers reliable over time. AI answers vary from run to run and with small wording changes, and platforms change their models and formats often. Each change means parser fixes and a new baseline.
Can I build on top of Spyglasses data?
Yes. The REST API lets you pull daily metrics and share of voice into your own tools, and the MCP server connects your data to Claude, ChatGPT and other AI assistants.
How much does Spyglasses cost?
Plans start at $29/mo for 25 prompts tracked daily on ChatGPT, Gemini and Google AI Overviews. Agency plans add white-label reports, a client portal and more prompts.
Why not just call the ChatGPT API every night?
You can, but API answers can differ from what people see in the ChatGPT app, and one run a night is a small sample. A tracker you can stand behind needs collection that matches what buyers see and enough runs to separate real change from noise.
Is Spyglasses the same as SEO SpyGlass or Visibility.ai?
No. Spyglasses (spyglasses.io) is an AI visibility platform. It isn't SEO SpyGlass, Link-Assistant's desktop backlink tool, and isn't related to Visibility.ai.
See what AI says about your brand
Start a free trial, add your prompts, and see your first daily results tomorrow.