Changelog
What shipped, and when
Dated notes on the changes you can actually see — new pages, new measurements, and the times we corrected something we had been counting wrongly.
Maintained by Anil Sayar, founder of Promvia · Last updated
Questions worth adding: buyer questions from Google's AI answers
Questions worth adding has a new section: buyer questions that Google already answers with an AI Overview, each with its monthly searches and the brands the answer names. It is refreshed monthly on every paid plan, for English-language and Turkish markets; Free sees a locked preview. How many searches each plan's refresh asks is on the pricing page.
Off-site plan: pages from outside your checks; suggested competitors
On Pro and Agency, the off-site plan now also lists pages Google's AI answers cite for your category and sites that link to your competitors but not to you, refreshed weekly and within the plan's page limit. On every plan, the Competitors page suggests businesses that answers to your own questions listed as a pick, with a link to their site.
AI Visibility Score: the GEO part no longer counts your citations a second time
The site-check score has two items about AI answers: whether they name you, and your share of voice. The Visibility Score already measures both, so its GEO part now uses only the site check's technical items. A technically sound site that AI does not name yet scores a little higher; a site AI already names scores a little lower. The API reports it as model version 3; do not compare scores across versions.
Off-site plan: the pages AI answers use, and how to reach them
A new page under Fix lists the third-party pages AI answers used for your tracked questions over the last 14 days, one row per page: who wrote it, what it lists and how to reach them. Promvia drafts the message; you send it yourself. How many pages and drafts each plan gets is on the pricing page.
Free: your questions are now checked automatically once a month
A Free site's tracked questions are now checked on Perplexity once a month, and the owner gets one email saying what changed since the previous check. It needs a confirmed email address and those emails switched on, and it pauses after three emails with no sign-in in between; signing in starts it again.
API: tooThin now reads the sample size alone
In the REST API, the MCP tools and the CSV export, tooThin (and the _sample_too_small columns) is now true when a rate was measured on fewer than 32 answers, whatever the rate. Before, it read the width of the observed interval, so more rates now report tooThin: true. The dashboard draws those rates as counts and a range, with no percentage.
New prices: Starter €49, Pro €179, Agency €499, with more tracked questions
Starter now tracks 25 questions (was 20), Pro 100 (was 80) and Agency 250 (was 220). Paying yearly is 10% off on every paid plan: €529, €1,933 and €5,389 a year. The free plan and the number of sites on each plan are unchanged.
xAI joins our sub-processors, for Grok
Grok, which you can add to your engines on a paid plan, runs through xAI's API. xAI is now listed in our data processing agreement and privacy policy: once you add Grok, it receives the questions you track there, never your account.
Your monthly checks now count per engine
From your next billing period, your plan's monthly allowance is counted per engine: choose fewer engines and you can track up to three times the questions, and automatic checks keep their own share that on-demand runs cannot spend. A billing period that started before this date finishes on the old shared pool.
Choosing your engines, explained
Help now has an article on choosing your engines: how to change them, when a change applies, what happens to your history, which engines cost the most to ask, and what daily checks cost on Starter. Pricing and the home page show the same worked example, so you can see what fewer engines buy before you sign up.
“Asked how often” is measured on Pro and Agency
How often a question is asked is now measured every Tuesday on Pro and Agency, and its band shows on the Citations page. Starter no longer includes it: a Starter account stops showing the band, and the question page says it is a Pro and Agency feature.
Checked daily on four engines, a weekly deep check on two
Pro and Agency now check every question daily on Perplexity, Gemini, AI Overviews and Google AI Mode, with a weekly deep check on ChatGPT and Claude. Won and lost alerts come from the daily checks too, still after two checks in a row. In Settings you can choose daily or weekly per engine; weekly tracks more questions.
A receipt by email for every paid invoice
Every paid invoice now sends the account owner one receipt by email, in the account's language: the amount, the period, the plan, the VAT lines as they appear on the invoice, and links to the invoice page and its PDF. An invoice with nothing to pay sends none.
Send a content draft to WordPress, as a draft
On Pro and Agency, a question's content draft can now be sent to WordPress as a draft post. Promvia only ever creates drafts: it never publishes, updates or deletes anything, and sending the same text again opens the draft already made.
“Asked how often” is now a band, not a Google number
On paid plans, how often a question is asked now shows as a band (rare, occasional or frequent) instead of a Google search-volume number. AI assistants publish no question counts, so a band from a proxy is the honest form. Each band names the term and the market it was measured on, and you can change the term.
Sentiment is now read by a model on paid plans
On paid plans, the tone of each AI answer that names you (positive, neutral or negative) is now read by a model instead of estimated from word lists. An answer that only repeats your name back from the question is not counted. The Free plan keeps the keyword estimate, and the column's tip says which one you are looking at.
AI crawler visits straight from your CDN logs
On Pro and Agency, AI crawler visits can now come straight from your CDN logs, through Cloudflare Logpush or a Vercel Log Drain; Starter keeps it until a billing period that started before this date ends. We keep only AI crawler lines, without the IP address, for 90 days, and a visit from a crawler's published IP range is marked verified. The Free plan no longer includes agent events.
Webhook run counts now count AI answers only
In the run-completed webhook, a question now counts as checked only when at least one AI answer engine answered it. A question where only Bing's search results came back now counts as failed, because a results page is not an AI answer. The same rule decides which questions can be announced as won or lost.
More tracked questions again: 20, 80 and 220, same prices
Starter now tracks 20 questions (was 15), Pro 80 (was 50) and Agency 220 (was 180), at the same prices. Every question is still asked on every surface your plan covers, and every AI answer is still sampled twice. The extra questions come out of the same monthly allowance as the checks you start yourself, so there is less room than before for repeated full re-runs on top of the weekly schedule.
More tracked questions on every plan, and Agency at €449
Starter now tracks 15 questions (was 10), Pro 50 (was 30) and Agency 180 (was 150), and Agency is €449 a month instead of €499. Starter's every-three-days trend checks cover the 10 questions you choose, the same way Pro's daily ones do. Starter and Pro prices did not change.
One name for the thing your plan counts
Pricing and the help pages now use a single term — tracked questions — for the unit every plan is measured in. No number changed; the wording did, because it used to switch names between the page that sells the plan and the page that explains it.
Availability and accessibility, in writing
Two new public pages. One says what we promise about uptime — including that there is no contractual service-level agreement yet, and why. The other says which parts of the site are known not to meet the accessibility standard we aim at, and where to report a barrier.
Tax is added at checkout, and every price now says so
Prices are quoted without tax, and the sentence saying so now appears wherever an amount is shown — the plan cards, the plan picker, the capacity packs and the pricing questions. The rate is worked out at checkout from your own country, and companies in the EU can enter a VAT number there.
See what is left this month before you run out
Your settings now show how many checks you can still start yourself this month, the date the month resets, and the fact that your automatic weekly checks hold a separate share your own runs can never spend. The same figures are readable through the API.
The weekly summary and the win/loss alerts are separate switches
They used to be one setting, so turning off the weekly summary also silenced the alerts that tell you an engine has started or stopped citing you. Each has its own switch now, and the account mail we cannot turn off — receipts, sign-in, password — is listed separately and labelled as such.
Yearly billing, and more capacity without changing plan
Every paid plan can now be paid yearly, and each tier states plainly what that does to the price rather than implying a discount it does not give. Pro and Agency can also add tracked questions or client sites in packs, so a growing account no longer has to jump a tier to fit.
The AI Source Index — which sources the engines really cite
A public panel of forty buyer questions is put to every engine we track, once a week, and every source those answers cite is published. Each week keeps its own permanent page, and a separate log records changes in the engines themselves.
A documented API, an MCP server and signed webhooks
Your visibility data is now reachable from your own tools: a read API with a published specification, an MCP server so AI agents can query it directly, and outbound webhooks that push results to your server instead of you asking for them. The developers page documents all three.
The site audit reads your whole site, not just the homepage
The deep scan starts from your sitemap and works outwards through the site, up to the page budget your plan allows. Findings now name the individual pages that carry each problem, so a fix has an address instead of being advice about the site in general.
A rebuilt home page
The front page was redrawn around a live product card and a band of things we can actually evidence — what the product measures, the public scoreboard, the written method. A young product has no customer quotes to show, so it shows the proof it does have.
First signals, for a site nothing cites yet
A new site used to open on a wall of zeros. The dashboard now shows the smaller wins that arrive first — being named in an answer without a link, or turning up for a question about your own brand — and says what each of them is actually worth.
An engine repeating your name is no longer counted as a mention
When the question itself names your brand, the answer usually repeats it without recommending you. Those echoes were being counted as mentions; they are not any more, and a question about yourself can no longer lift your own visibility score.
A feed for the guides, and rivals on the public scoreboard
The guides publish a feed you can subscribe to, in both languages. On the public scoreboard, every question we are not mentioned for now names the competitors the engine recommended in our place.
Every public page is served from cache again
The language moved into the address, and the theme and signed-in state moved into the browser. Marketing pages, guides and free tools are now prepared ahead of time instead of being built for each visitor: they open faster, and both languages can be crawled.
Promvia measured on itself, in public
A public scoreboard shows what the answer engines currently say for our own tracked questions, taken from the product's own checks rather than written by hand. The questions we lose are on it too — that is the reason for publishing it.
Two more free tools, with no signup
A citability check scores any homepage on the citation-readiness signals we can verify. A head-to-head runs the same scoring against a competitor and names exactly which signals they have and you lack.
Turkish pages have addresses of their own
The language used to be remembered in a browser setting, which meant a search engine only ever saw one version of the site. Every public page now has its own Turkish address, and the guides were written in Turkish to match.
A library of guides
Long-form guides on getting recommended by AI assistants: what the engines look for, how to measure what they send you, and where the honest limits are. Each one opens with a direct answer and names its sources. The library has kept growing since.
Only changes a customer can see are listed here. A great deal of the work never surfaces; that is a good sign rather than an omission.