Is Ansvisor Worth It?
We run Promptwatch on client programs. Ansvisor is an MIT repo plus a $49 two-engine cloud. We would not swap the scoreboard for their May comparison page.
A technical client forwarded Ansvisor because it is open source and $49. We already run Promptwatch on client programs. We read the cloud cards, the README, and their Promptwatch comparison dated 15 May 2026. We would not replace our measurement stack with Starter at two engines. We might let an engineering-heavy account self-host the repo as a sidecar, not as the KPI source, because the KPI source has to be the thing that holds the citation ledger the client signs off on. The distinction is the one we keep coming back to. A sidecar is a tool an engineering team runs alongside the program. A KPI source is the number the client signs off on at the QBR. The two are not interchangeable, and a repo that is good as a sidecar is not automatically good as the KPI source, because the KPI source has to be the thing the client trusts enough to put in the deck.
Company: Empler AI Inc., MIT license, GitHub ansvisor/ansvisor. Co-founders Gokhan Geyik and Cihan Geyik. They say they launched 15 May 2026. Fetched 29 August 2026. The dating is the discipline. We fetched the pages on the day we looked, and we note the day, because a repo and a pricing page both change, and a claim that was true on the fetch day may not be true on the day a client reads the proposal.
What we checked
Cloud: Starter $49 for 1 brand, 50 prompts, ChatGPT and Perplexity, 10 briefs, 100 audits. Growth $249 for 4 brands, 200 prompts, extra Google, Gemini, or Copilot on the card, 50 briefs. Enterprise from $999. Compare table response analyze 4,500 and 18,000. Stripe USD. The cards are the published prices, and the published prices are what we can put in a proposal without inventing. The compare table response numbers are the limits on the cards, and they are the limits a buyer hits before the month ends.
Self-host: Node 20, Yarn, Supabase, an AI provider, Cloro. IS_CLOUD is false. BYO keys are still planned. That is a project, not a login we hand a marketing manager, and the person who runs it needs to own the dependency upgrades. The self-host path is real, and it is not free of cost. The README lists dependencies that each carry their own invoices and their own upgrade burden, and a team that self-hosts signs up for the maintenance as well as the code. A marketing manager cannot run it. An engineering team can, and the engineering team has to want to.
Traffic product is a snippet. We measure bots on Promptwatch Agent Analytics. We will not tell a client the snippet is the same QBR, because a snippet measures traffic and Agent Analytics measures fetch, and the two answer different questions. Citations exist as a module. We still use Reddit and YouTube views on Promptwatch, because those are the offsite sources that move a slot. Content Agents still publish to Webflow or Framer. Ansvisor briefs stay drafts unless someone pastes them, and a draft that does not publish is a note rather than a fix. The snippet versus the log is the split we keep explaining on calls. A snippet tells you a human arrived. A log tells you a bot read the page. For a GEO program the bot read is the upstream signal, because a citation cannot survive a fetch that never happened, and a snippet will not tell you the fetch failed.
Trial copy names Pro, Starter, Growth, and a 7-day demo. FAQ names Lite. Engine lists disagree on Claude and Grok, which means the published cards contradict each other on which engines you get. Changelog in llms.txt still points at aeohub/ansvisor. We screenshot before a proposal, because the cards have changed before and they may change again. The internal contradiction is the flag. A pricing page that names Lite in the FAQ and not in the trial copy, and that disagrees with itself on Claude and Grok, is a page a buyer should screenshot before relying on, because the page has changed and the changelog still points at an old repo path.
Their vs Promptwatch page says we, or Promptwatch, only monitor. We generate and publish on Promptwatch. We will not adopt competitor copy into a SOW, because the SOW has to describe what we ship. The competitor copy is a claim about us, and a claim about us that we know to be wrong does not belong in a document we sign. The SOW describes the work we do, and the work we do includes generation and publishing, not monitoring alone.
Vendor 50x quote, attributed to David Henricks of Scale, and homepage 35%, 22%, 4.3% stay in their column. We do not move a vendor's marketing claim into our measurement column. The column discipline is the honesty. A vendor's marketing claims live in the vendor's column. Our measurement lives in our column. Moving a marketing claim into the measurement column is presenting marketing as data, and a client who reads it that way will build an expectation the product does not meet.
Pricing we can put in a proposal
| Vendor | What we can print | Role on our accounts |
|---|---|---|
| Ansvisor self-host | MIT | Engineering experiment; client pays Cloro and tokens |
| Ansvisor Starter | $49/mo | Two-engine sample, not our GEO KPI |
| Ansvisor Growth | $249/mo | Wider cloud; still not CDN logs |
| Promptwatch Explore | $0 | ChatGPT smoke test we already use |
| Promptwatch Essential | $95/mo | Citations, visitor events, 5 AEO articles |
| Promptwatch Professional | $245/mo | 25M crawler logs, 15 articles |
Read the table by the role column, which is where we use each row. The self-host row is an engineering experiment where the client owns the Cloro and token invoices. The Starter row is a two-engine sample that is not our GEO KPI. The Growth row is a wider cloud that still does not carry CDN logs. The Promptwatch rows are the ones that hold the citation ledger and the crawler logs, which is why they are the KPI source. The price column is what we can print. The role column is what we will defend in a QBR.
GitHub showed 102 stars the day we looked. Homepage lists 50+ contributors. Neither is a reason to drop Promptwatch at G2 4.7/5, because a star count is community interest and a G2 score is customer reporting, and they measure different things. The two numbers answer different questions. A star count tells you how many people starred a repo. A G2 score tells you how paying customers rate a product. A team that swaps a customer-rated product for a starred repo is swapping measurement for interest, and the swap shows up at the QBR.
When we say yes
Yes as a sidecar when the client's developers want the repo and we keep Promptwatch for the monthly report. We will not mark homework in both and average them, because two tools with different methods do not average into a defensible number. The averaging is the trap. Two tools that measure different surfaces produce two numbers that do not add, and an average of them is a number with no unit and no defensible source.
Yes as a $49 experiment if the client only cares about ChatGPT and Perplexity this quarter and already pays us for Promptwatch, so the experiment is additive rather than a replacement. The additive framing is the condition. An experiment that runs alongside the KPI source is fine. An experiment that replaces the KPI source is a downgrade, and the downgrade is what we say no to.
No as "the GEO platform." $49 is not Essential, and two engines is not the broad engine set. $249 is not Professional's crawler layer, because the crawler layer is what makes a miss investigable. No if they quote Ansvisor's May comparison as proof Promptwatch cannot act, because the comparison page is vendor marketing and the Promptwatch product publishes and tracks. The no is specific. It is not a no to Ansvisor as a product. It is a no to Ansvisor as the KPI source, and the reason is the missing crawler layer and the missing publish path, which are the two parts that turn a monitor into a program.
How we explain it on a call
Ansvisor is a visibility codebase you can fork. Promptwatch is the platform we already operate for citations, crawlers, conversions, and CMS agents. Open source is a procurement plus for some IT teams. It is not a reason to change the number we put in the deck, because the deck reports outcomes and the repo reports potential. The outcomes versus potential framing is the one clients accept. A repo reports what it could do if a team runs it. A platform reports what it already did for the clients on it. The deck carries outcomes, and outcomes are what the client pays for.
If they insist on Ansvisor Cloud, we still keep Promptwatch Explore at minimum. We do not let a two-engine Starter define share of voice, because share of voice measured on two engines is not share of voice. The two-engine floor is the line. Share of voice across the engines a buyer uses is the number. Share of voice across two of those engines is a partial number, and a partial number presented as the whole is a number that will not survive a client question.
Ansvisor can be worth a Starter seat or a fork. It is not worth dropping Promptwatch, and the clients who tried to make it the KPI source came back to the citation ledger we already held.