Your buyers ask AI. Does it name you?
SignalKit asks seven AI platforms the questions your buyers ask, shows who they name and which sources they trust, explains why your numbers moved, and turns the gaps into pages you can publish.
US$3 per question per region per month, every platform included, no per-seat fee and no minimum.
What changed
Mention rate rose 9.1 pts over 90 days, compared like-for-like on 22 of 24 prompts. ChatGPT accounts for 4.8 of them. g2.com started citing you on five prompts.
How a move is broken downFictional brands and invented figures at the real layout. Not customer data and not an industry benchmark.
AI already answers your customers. It may not name you.
Ask an assistant which tool to buy and it names a handful of brands, in an order, with a few sources behind them. That is a ranking, and nobody publishes it. SignalKit asks the questions your customers ask and keeps every answer, so the ranking becomes numbers you can watch.
- Named
- whether the answer names you, and which rivals it names instead.
- Position
- where in the answer you first appear, when you do.
- Cited
- whether a source the answer used is your own domain.
What's the best expense tool for a 15-person startup?
For a team that size, most finance leads shortlist CostGuard for its card controls, PayCadence if approval flows matter most, and InvoiceIQ for receipt capture. All three connect to the common accounting tools.
Sourcesg2.comcostguard.comcapterra.com
Seven AI platforms. Not a sample of one.
Answers diverge sharply between platforms. The brand that leads in ChatGPT often trails in Perplexity, so checking one and calling it “AI visibility” is how you get a confident, wrong answer.
- ChatGPT41.0%Daily
- Claude44.8%Daily
- Gemini36.3%Daily
- Perplexity29.5%Daily
- DeepSeek26.4%Daily
- Grok31.8%Mon and Thu
- Google AI Overviews33.1%Mon and Thu
Illustrative figures. Mention rate — the share of eligible answers that name the brand — for one fictional brand over 234 answers per platform. Not customer data, and not a benchmark.
Every question can use all seven at no extra cost. Five are asked every day; Grok and Google AI Overviews are asked twice a week, on Mondays and Thursdays, so their figures rest on fewer answers.
When a number moves, the reason is one click away
Every change on the dashboard opens the same breakdown: which platforms, prompts and topics moved it, and by how many points each. It is how a weekly number becomes a decision rather than a guess.
Change breakdown
IllustrativeMention rate · last 30 days vs the 30 beforeCompared like-for-like on 22 of 24 prompts and 6 of 7 platforms.
Grok returned 18 of 204 expected answers in 30 days, under half its schedule, so it is left out of this comparison.
By platform · also by prompt, topic, competitor and source
| Platform | Now | Change | Contribution |
|---|---|---|---|
| Gemini150 answers | 22.0%was 29.3% | −7.3 pts | −1.6 pts |
| ChatGPT164 answers | 34.1%was 35.0% | −0.9 pts | −0.6 pts |
| Perplexity139 answers | 30.2%was 31.0% | −0.8 pts | −0.2 pts |
| Claude148 answers | 36.5%was 35.1% | +1.4 pts | +0.3 pts |
| Google AI Overviews59 answers | 28.8%was 28.2% | +0.6 pts | +0.1 pts |
| DeepSeek142 answers | 31.0%was 31.0% | 0.0 pts | 0.0 pts |
| Mix and rounding | +0.1 pts |
Mention rate fell 1.9 pts. Gemini accounts for 1.6 pts of it.
- The rows add up. Every row is folded from the same answers as the headline, so the contributions and a small residual are exactly the move. Split it by platform, prompt or topic; competitors and sources are shown as counts beside it.
- Like for like, or not at all. When a prompt was added or a platform stopped answering, the change is taken on what both periods measured, and the panel says which. With under 70% of this period's answers in common, or a change in how answers are captured, the change is withheld with its reason instead of guessed.
- Coverage is a signal. A platform that returns under 50% of the answers its schedule expected is named on the panel, left out of the comparison, and raised as a data-health alert with expected against received.
- The sentence is free; prose is priced. The line above the rows is arithmetic over them and costs nothing. Explain buys one model paragraph over the same rows, shows its price first, and is thrown away if it quotes a figure the rows do not hold. 20 change explanations per 30 days are included.
From a missing mention to a published page
Measurement shows where you are absent. To do ranks what to fix every Monday, Content drafts the page in your brand's voice, a person approves it, it goes live on your site, and To do then shows which platforms cite it.
1Measure
Prompts › Answers
best expense tool for startups
- ChatGPTNot named
- GeminiNamed · Position 3
- PerplexityNot named
Every tracked prompt on every platform you choose, stored with who was named, in what order, and which sources were cited.
2Decide
To do
OwnedNewHigh impactPublish a comparison page for expense tools
CostGuard is named ahead of you on 3 prompts.
Create brief
Each Monday the weekly brief ranks what to do by prompts affected, search demand and the size of the gap, and links the measurement behind each item.
3Write and publish
Content › Pipeline
Expense tools compared
Plan: Prompt coverage opportunity
- Review
- Approved
- Draft
- 4 units, critic on
- Publishes to
- Shopify blog
Planned from the evidence, drafted from your Brand, scored and critiqued, and published only after a person approves it: a pull request to your Git repository, a Shopify blog post, or copied out by hand.
4See what changed
To do › Done
Cited by 2 of 7 observed platforms after 12 days.
Target prompt, 14 days before and after
28.6%to42.9%+14.3 ptsDone items with a published piece show the platforms citing it and its target prompts, 14 days before and after publication. They make no claim about cause.
What you pay for which partTo do and its weekly brief come with monitoring. Drafting and publishing are Content & Growth, a separate subscription: managed models, or your own OpenAI, Anthropic or Z.AI key, paid in allowance units where one unit is one model call. Setting up your Brand, its materials and your destinations costs nothing, and saved drafts stay readable if you stop.
- Suggestions that keep arriving. New prompts proposed from the searches the platforms ran behind your answers, from answers that named a rival and not you, and from your Search Console queries. Nothing is billed until you add one, and the cost is shown first.
- Segments. Split prompts by the markets you compete in, each with its own prompts and competitors, and read every page inside one.
- Site audits. A free quick audit any time and 2 deep audits every 30 days per project. A failing check whose fix is a file — robots.txt, llms.txt, a schema snippet — is handed over as that file.
- Products. Bring a Shopify, Merchant Center or CSV catalogue into a segment, see which products the answers name, and get an alert when an answer quotes a price your catalogue does not. Matching answers to products is metered: 2 runs per 30 days.
One price per question.
US$3per question · per region · per month
20 questions in one region is US$60 a month, whichever platforms answer them. Writing and publishing pages is Content & Growth, a separate subscription whose price Billing shows before you confirm.
- All seven AI platforms on every question
- No seats, tiers or minimum spend
- $6 one-time invoice credit when you sign up
- REST API and MCP connector included
Start with a free audit of your site.
Paste any URL and see whether your site is ready for AI to read it, scored out of 100. It reads your pages, robots.txt, sitemap and structured data — it asks no AI platform about you. No signup required.
Frequently asked questions
Pricing, platforms, what is measured, and what is not.
Standard Monitoring costs US$3 per tracked question per region per month, with all seven platforms available at no extra cost. Projects, exports and API access are included, and nothing is billed per seat. Add a question mid-month and you pay the prorated remainder; remove one and the charge stops. Enhanced measurement is an optional US$10 per active prompt-region monthly add-on for two samples per scheduled run and Claude web search. Content & Growth is a separate optional subscription with Managed and bring-your-own-key plans.
New accounts receive $6 in one-time credit toward Stripe invoices. A card is required before tracking starts, and every active question costs US$3 per region per month. Unused credit carries forward; normal billing continues once it is used. There is no recurring free prompt quota. Where Content & Growth offers a trial, it is once per account and card-backed, and renews as a paid plan unless cancelled; Billing shows the plan, the trial length and units, and the renewal price before checkout. The website audit is free and needs no account.
SignalKit tracks seven: ChatGPT, Claude, Perplexity, Gemini, Google AI Overviews, Grok and DeepSeek. You can select any of them for a tracked question at no extra cost, because a brand that leads on one platform often trails on another and a single platform is not a sample.
SignalKit uses provider APIs, not consumer chat UIs. ChatGPT, Claude, Gemini and Grok are called through Cloudflare's AI Gateway against each vendor's own API; DeepSeek is a Cloudflare-hosted model; Perplexity is called directly; and Google AI Overviews come from SerpAPI. These are API measurements, so they are close to but not identical to what a person sees in the consumer app. Live provider output can vary between runs, so SignalKit timestamps and stores each response instead of describing it as deterministic.
Every account can create API keys for the supported REST endpoints. Pull mention and cited rates, sentiment, position data, citations and raw AI responses programmatically. Keys can be scoped to a single project, which is useful if you want to give a client a feed of their own data and nothing else.
SignalKit hosts an MCP (Model Context Protocol) connector at https://app.signalkit.ai/api/mcp. Add it as a remote MCP server in any client that takes a URL — Claude, Claude Code, Cursor, Cline — and authenticate with OAuth or a SignalKit API key. It covers projects, monitoring, Content and Growth. The app-operation tools preserve the app's permissions and paid-feature gates: signed in with OAuth, an agent can do what your role allows, and billing changes also need the owner's permission in the app. Over an API key, human-only actions return a link to the app. The separate stdio package exists in our source tree but is not published to npm.
Yes. API keys, structured JSON responses and the MCP server make it possible to query brand monitoring data from scripts, CI/CD or agent frameworks.
LLM brand monitoring tracks how AI assistants like ChatGPT, Claude, and Perplexity mention your brand when users ask questions. As more people use AI for product research and recommendations, knowing when and how AI talks about your brand is essential. This is sometimes called Generative Engine Optimization (GEO) or Answer Engine Optimization (AEO).
Your first results appear within minutes of adding prompts: creating one dispatches it straight away rather than waiting for the next scheduled run. After that the scheduler runs at 04:00 UTC, and cadence differs by platform — five run daily, while Grok and Google AI Overviews run on Mondays and Thursdays. Standard measurement takes one sample per scheduled run. The optional Enhanced measurement add-on takes two and enables Claude web search. Results update in your dashboard and API as they land.
Price, access and transparency. Monitoring costs US$3 per tracked question per region per month, so the bill is arithmetic you can do in your head before you sign up. Every account gets API access, the MCP connector, all seven platforms and up to 999 projects, and nothing is billed per seat. We also show the raw AI responses, so you can check exactly what each platform said rather than taking a score on trust. Where we are behind: there is no SSO and no procurement track, and sentiment is scored from a keyword list rather than by a model. We do not publish benchmarks or industry averages, because we have no consented dataset to build one from.
Mention rate is the share of eligible answers that name your brand: the numerator is the answers naming you, the denominator is the answers we asked for and received in the period. Cited rate is the share of answers whose sources were observable that carry a supported citation to your own domain. New Perplexity Agent measurements distinguish references the answer used from pages it only retrieved; older Perplexity Sonar answers remain not observable because that mapping was not retained. Grok and DeepSeek are observable only when their response carries traceable source links. They are reported separately and they move separately — an answer can recommend you without linking to you, and cite you without naming you as a recommendation. Both are shown with their denominator, so a rate is never a number without a population behind it.
It is made like for like: on the prompts, platforms and regions both periods measured, and the change says so — for example 22 of 24 prompts and 6 of 7 platforms. A platform whose answers fell by more than half is left out and named rather than averaged in, and the change breakdown shows how many points each platform, prompt and topic contributed. When under 70% of this period's answers are in common, or the way answers are captured changed, the change is withheld with its reason instead of shown.
It is the half of SignalKit that writes. To do ranks what to fix from your measurements each week; Content turns it into a plan, drafts pages in your brand's voice from the Brand profile and materials you approve, scores and critiques them, and publishes to a Shopify blog or a Git repository behind an Astro site once a person approves, or leaves the approved text for you to copy out. After publication it reports which platforms cite the page and how its target prompts moved, without claiming why. Drafting and publishing are a separate subscription from monitoring, with managed models or your own OpenAI, Anthropic or Z.AI key, paid in allowance units where one unit is one model call. Setting up your Brand, its materials and your destinations is free, and the current price is shown on your billing page before you confirm.
Yes, and there are two separate settings. The first is the company role: an Owner or an Admin reaches every project in the company, including ones created later, and manages billing, members and API keys; a Member has no company-wide authority. The second is the per-project grant a Member gets — Viewer reads the project and exports what is stored, Editor can also change prompts, brands and settings. Owners and Admins hold no grants because their authority already covers every project. You invite by email and choose the project grants at the same time; they apply when the invitation is accepted. Members are not billed: seats cost nothing, Monitoring is billed by active prompt-regions, while optional Content & Growth has a separate subscription. The maximum number of members comes from the plan configuration on the identity provider rather than from a number we set.
Yes. SignalKit stores the full raw response from each platform query. You can verify exactly what ChatGPT, Claude, or Perplexity said about your brand, see which competitors were mentioned, and check which sources were cited. Citation quality differs by platform: ChatGPT, Gemini, Perplexity and Google AI Overviews return structured source lists. Claude does too when Enhanced measurement enables its web search; standard Claude, Grok and DeepSeek otherwise expose only URLs present in the answer text.
One Monitoring subscription covers the account. Its standard quantity is the number of active prompt-region pairs in switched-on projects: one question in two regions is two units, at US$3 per unit per month. All projects and domains on your account contribute to the same subscription. Adding a question raises the quantity; archiving one, or switching its project off, lowers it. Mid-month changes are prorated on your next invoice, and the billing page breaks down the monthly rate by project. Enhanced measurement is an optional US$10 per active prompt-region add-on on the same Monitoring subscription. It can be enabled per project for two samples per scheduled run and Claude web search. There is no prepaid balance or per-platform surcharge, and each question can use any of the seven platforms at the standard price. AI assistance and deep audits have included usage limits shown on the pricing page. A card is needed before your first question starts running. Content & Growth is a separate subscription on the same billing account and may renew on a different date.