Changed — how AI visibility is measured

"AI selection rate" is now "citation presence", and it is measured differently. The old number counted how many AI models said they mentioned you, divided by how many checks we had stored. Two things were wrong with that. The model was answering a question about itself, which it can get wrong in both directions, and a check that failed for our own reasons — a provider timeout, a rate limit — simply left no record, so an outage on our side was indistinguishable from an answer that left you out.

Citation presence is read from the sources an AI answer actually linked to, wherever the provider gives us them, and every number now says which way it was measured. Where a model still gives us nothing but its own account, the label says so.

Your number may move, up or down, without anything about your site having changed. That is the measurement being corrected, not your visibility.
Every rate now shows what it was divided by. How many answers it is based on, how many attempts never came back, and — where an answer listed its sources in order — how often you were the first source cited rather than merely present.
What a flat reading actually rules out. A rate built on a handful of answers cannot detect a small change, so we now say how big a change this much evidence could have spotted, instead of implying the number is steady.
Failed checks are visible instead of silent. An attempt that never produced an answer is recorded and shown, and it is left out of your rate rather than counted as "not cited".
Measurement changes break the line. When we change a model or fix how citations are read, the chart marks the date and says that readings either side of it are not directly comparable.

Changed — AI visibility cadence and pricing

Each tracked keyword's first prompt is now checked for free, every week, on every AI engine enabled in your plan. The engines that used to cost extra credits on their own are part of that included check now, and the features page always shows the current engine list. On a paid plan, turning AI visibility on for a keyword no longer spends credits by itself.
AI visibility checks now run once a week instead of every night. AI answers shift over weeks, not hours, and a weekly check with more engines tells you more than a nightly check with fewer. Each keyword keeps its own fixed weekday.
Extra prompts cost credits. A keyword can have up to three prompts. The first one is included; each extra prompt uses credits every week it runs, and we show the price and ask you to confirm before you add one. If your balance runs out, extra prompts are skipped until you top up, and the first prompt keeps running.
"Run AI visibility check" still uses credits. It runs every prompt right away and uses the same per-prompt price for each one, the first prompt included, because it is an extra check on top of the weekly one.

Credits you already spent on nightly checks are not affected, and those past entries still show in your usage breakdown under their old names.

Changed — scoring

llms.txt, ai.txt, TDMRep and FAQ/Speakable markup no longer affect any score. They are still checked and still shown, under their own heading, but they no longer move your CITED Score, your GEO readiness score, or the citability sub-score.

The reason is that we could not show they do anything. No AI answer engine has confirmed reading llms.txt, and the published log studies — Ahrefs, and SE Ranking across 300,000 domains — found roughly 97% of llms.txt files receive zero AI-crawler requests and no measurable citation effect. ai.txt is a competing draft with no ratified spec. TDMRep is a W3C Community Group report with no confirmed crawler adoption. FAQ rich results were retired from general search in 2023 and Speakable was a narrow, discontinued voice feature; neither has a demonstrated effect on being cited.

Between them they carried 45% of the technical-accessibility score and 40% of the citability sub-score. That meant a customer could raise those numbers substantially by publishing text files, without changing anything an AI engine can actually see — so the score was rewarding the wrong work.

We removed them rather than reducing their weight. A smaller weight would still be a claim that publishing the file helps, and we have no evidence for that claim at any size. Structured-data validity and completeness checks are unaffected: invalid markup is still a real defect and is still reported.
A number on your page only counts as evidence when it is sourced, dated and on topic. The GEO readiness check used to reward how many numbers a page contained. That rewarded the wrong thing: ten unattributed percentages scored better than one figure with a named source. A statistic now counts when the same passage says where it came from (a link, "according to", a named study or survey, or how it was measured), gives a year, and sits under a heading about the same topic. A page passes when at least half of its statistics meet that bar.

This follows the published evidence, which supports relevant, accurate statistics being cited more often and gives no support to a count target.
Attributed quotes from named people now earn a small amount of citability. A quotation with a name and, ideally, a role is something an AI answer can lift and attribute. The weight is deliberately small: the evidence for it is moderate, and a pattern match cannot tell whether the quote says anything useful.

Your CITED and GEO readiness scores may move, up or down, as a result of all three changes above. Nothing on your site changed; what we are willing to claim, and what we count as evidence, did.

Added — AI visibility

Four more AI engines tracked. Google AI Mode, Grok, Meta AI and DeepSeek join the engines checked for your keywords, and each engine's reading says whether its citations come from real linked sources or only from the model's own account.
Why you lost a citation, ranked. When you stop being cited for a keyword, we now put the likely explanations in order — your page changed, a competitor's did, the answer's sources moved, or the engine itself did — group repeat losses into one incident instead of alerting every week, and hold back when the cause was an outage on our side. Each incident shows what changed on the page between the two readings.
When an AI crawler last fetched the page. Where you have uploaded server logs, a citation loss shows the last time each AI crawler actually fetched the page in question, so "they cannot see it" and "they saw it and chose someone else" stop looking the same.
A weekly check that AI bots can reach your pages. We probe your pages as the AI crawlers see them, per host, reading each host's own robots.txt, and record the page state we found.
"Cited instead of you." The actual pages AI answers cite for your keywords, sorted by how often, and grouped into what you can do about each one — ask a round-up to include you, claim a review profile, pitch a publisher, or read a competitor's page properly. Classified from the page address, so it costs nothing and is occasionally wrong; the link is right there to check.
Sub-query coverage on the AI tab. AI answers break a question into narrower searches. You can now pick the ones we inferred from a keyword's AI answer and track them as keywords of their own. Each uses one keyword slot on your plan, like any keyword, with no extra charge. For each keyword you then see how many of its sub-queries you rank in the top 10 for, and whether the page that ranks has a heading that answers each one. The match compares words, not meaning, and says so: "integrations" and "works with Slack" will not match.
Freshness in context, per AI engine. An out-of-date page is still flagged the same way, and now it also tells you what we know about how each engine treats freshness, instead of one rule for all of them.
Agent-readiness checks. Scans now flag content that only appears on hover, list endpoints an agent cannot page through, and whether your site offers a WebMCP endpoint.

Added — AI traffic you can actually count

See AI assistant visits in your own analytics. Connect Plausible or Google Analytics 4 and KinetixSEO reads the visits that came from ChatGPT, Perplexity, Claude, Copilot and the rest, so AI referrals sit next to your search traffic instead of being invisible.
AI referral traffic from uploaded server logs. No analytics connection needed — upload access logs and we count the same visits from them, alongside the crawler hits already shown.
Which of your sitemap pages AI crawlers actually fetched. Uploaded logs are matched against your sitemap, so you can see the pages no AI crawler has ever requested.
A free AI crawler log analysis tool. Public page, no account needed.

Added — elsewhere

Connect KinetixSEO to Claude and ChatGPT in one click. The MCP server now supports OAuth, so Claude and ChatGPT's connector can sign in properly instead of needing a hand-made API key, and every tool shows its real name and its live credit price before it spends anything.
Crawl your site from the CITED Score tab. A button to run the crawl the CITED checks need, with a four-hour cooldown, plus a legend and a score summary on the tab so each check reads as something you can act on.
Branded search share in your search insights. How much of your Search Console traffic comes from people already searching your brand name, in the stat strip.
The free check now explains itself. Every score on the public check says which signals built it.
Keyword opportunities from search results we already fetched. Rank tracking discards everything below your position; we now keep it, cluster keywords whose search results overlap, and turn that into ranked opportunities and content briefs at no extra cost to you.

Changed — plans and monitoring

Downgrading no longer silently over-runs your new limits. Sites, keywords and competitors above your new plan's limits are paused rather than deleted, and you choose which to un-pause.
Monitoring stops when a plan lapses, and restarts when it resumes. Nightly work no longer runs for a lapsed account. Credits you paid for are never deleted.
Keyword opportunity checks run weekly instead of nightly, per site, which is the cadence the data actually changes at.

Fixed

Yearly subscribers' monthly credits. A yearly plan now refills its plan credits every month, and comped accounts renew at no charge so their credits refill too.
"Citation volatility" is now "Citation presence change", and it only fires on a real change. The old alert reacted to day-to-day churn in which sites AI answers cited, and with a handful of prompts that churn is mostly noise. The new alert compares your citation presence prompt by prompt against the previous two weeks, per AI engine, and only tells you when the change is statistically clear, at least 10 percentage points, and confirmed on two separate check dates. When we change how an engine is measured, the alert waits rather than reporting the measurement change as a real one.
Page-change alerts for a removed or rewritten H1 and a changed H2 structure now actually fire. A mismatch in how heading levels were read meant these checks never found a heading, so they could never alert. You may start seeing them after your page's next scan, once there is a new snapshot to compare against.
The search-experience (SXO) check now reads your page's H1 instead of always falling back to your page title, which is the better description of what a page is about. Your SXO result can change as a result.
A clean brand no longer reads as an unmeasured one. Brand mention checks now record that they ran, and say when the weekly check happens.
Rank tracking tells you when a night went missing. You get an alert if a night of rank snapshots does not land, provider timeouts are retried instead of failing outright, and a malformed result no longer breaks the run.
Rank-tracker traffic no longer shows up as your own opportunity. Our own position checks were being counted as demand; they are excluded, and the heat map now marks cells no AI engine is asked about as "n/a" instead of leaving them blank.
Bing Copilot checks no longer fail on a provider response we were misreading as an error.
Larger sites' sitemaps are fully walked. The whole sitemap tree is crawled under one fetch budget, and that budget scales with your plan instead of stopping at the first index file.
Your domain is identified correctly. Registrable domains are resolved against the real Public Suffix List, so a country or multi-part suffix is no longer mistaken for someone else's site.
Free crawl samples no longer trigger paid AI analysis, and admin recrawls re-analyse the pages actually sampled.
Dark mode and mobile. The mobile menu is readable and scrollable, the desktop footer columns are back, pipeline stage badges and the Sites pages use theme colours instead of hardcoded light ones, and the real contrast failures on the homepage and footer are fixed.
Charts. One legend under the visibility chart instead of two, a readable axis, and the CITED radar is drawn in SVG so the Sites page no longer loads a charting library at all.
The file picker on log and CSV uploads is visible again.
Security and reliability. Stricter DNS handling on scans, API key listings scoped correctly, an interrupted llms.txt job no longer leaks a queue error to visitors, and unreachable sites name the real cause.
Site SEO. Every static page's title and meta description now fits within the length limits, the sitemap sends a public cache header, /index.php redirects to the site root, and the pricing markup carries a high price.

Citation numbers you can check, weekly AI visibility for every keyword, and AI traffic in your analytics