Key takeaways
- A page has a life cycle: draft, staged, published, crawled, cited, clicked. Most teams only see the first two stages and guess at the rest.
- Each stage needs its own signal. Crawler logs show whether bots read the page, citation tracking shows whether AI used it, and visitor analytics show whether anyone clicked.
- Citations and traffic are different layers. Aleyda Solis's April 2026 analysis found 57.7% of AI traffic landed on homepages or brand-entry pages, while only 3.0% of citations pointed there.
- Diagnose by stage. A page that is not crawled has an access problem. A page that is crawled but not cited has a content problem.
- Cited pages decay at different rates, so tracking has to continue after the first citation.
Why a published page is not a finished page
Most content workflows end at the publish button. The CMS says "published", the ticket closes, and everyone moves on. Then three weeks later someone asks whether the page is showing up in ChatGPT, and nobody knows.
The gap is that "published" is a CMS status, not an outcome. For AI search, a page can be live and still invisible for at least four different reasons: the bot never fetched it, the bot fetched it and got an error, the bot read it but the engine never retrieved it for any tracked prompt, or it got cited and nobody clicked.
Tracking a page through all of these stages is what I mean by following document status along the live URL. You keep one record per page and attach every later event to it.
The stages a URL goes through
| Stage | What happens | Signal that confirms it | Typical failure |
|---|---|---|---|
| Draft | Page is written, not public | CMS status | Brief ignores real prompts |
| Staged | Page is reviewed or scheduled | CMS status, publish window | Sits in review for weeks |
| Published | Live URL exists | HTTP 200, in sitemap | Orphan page, no internal links |
| Crawled | AI bots fetch it | IP-verified crawler logs | Blocked by robots.txt, CDN or WAF |
| Cited | An AI answer lists it as a source | Citation tracking per prompt | Crawled but never retrieved |
| Clicked | A user or agent lands on it | AI-referred visitor analytics | Cited without clicks |
The useful part is that each stage points to a different fix. People often treat "we are not getting cited" as one problem. It is at least four.
Stage 1 and 2: draft and staged
Tracking starts before the page exists. A draft should be tied to a reason: a prompt, a topic, or a content gap you can name. If you cannot say which AI answers the page is meant to appear in, you are guessing.
Two habits help here.
- Write the brief from real prompt data, including the query fan-outs the engine runs. Promptwatch Data's ChatGPT query fanout report tracks how many web searches ChatGPT triggers per response, which is a good reminder that one prompt becomes many searches.
- Record the status transitions. Draft to staged to published to live URL. When a page is later cited, you want to know which version and which date it came from.
This is where a platform that connects publishing to measurement earns its keep. Promptwatch tracks every document's draft, staged, published, and live URL when you publish through its WordPress plugin, Webflow integration, or Framer, then follows the live page through AI crawls, citations, and AI traffic.

If you track status by hand, a spreadsheet with one row per URL works. It gets painful past a few dozen pages.
Stage 3: published but is the URL healthy
Before blaming the AI, check the boring things.
- Does the URL return 200 and render without JavaScript tricks that hide the content?
- Is it in the sitemap?
- Does it have internal links pointing at it? Orphan pages are common after a batch publish.
- Are title, description and headings present and not duplicated?
A weekly site audit catches most of this. Promptwatch runs 21 checks weekly, including orphaned pages and pages missing from the sitemap. Screaming Frog, Sitebulb and Lumar do similar jobs if you already use them.

One tip: submit new or updated pages for indexing right after you publish. It costs nothing and shortens the wait for the first crawl.
Stage 4: crawled, the stage most teams skip
This is the stage I would add to any reporting setup first. Crawler logs tell you whether AI bots actually fetched the page, how often, and what happened when they did.
Analytics tools do not show this, because bots do not run your JavaScript tag. You need server or CDN logs.
Promptwatch Data's AI crawler traffic report covers IP-verified requests from June 8 to September 6, 2026. OpenAI made up 79.8% of verified AI crawler requests in the week of August 31 to September 6, down from 94.8% in the week of June 8 to 14. The mix is shifting, so a robots.txt rule written last year may not cover the bots that matter now.
The report also separates training crawlers from answer-time fetchers. They are different bots with different jobs, and blocking one does not mean you block the other. Its practical advice is simple: compare your own logs with the report. A provider that is big in the report but near zero in your logs usually means you are blocked or unreachable for that provider.
Two more points on bots:
- Verify by IP range, not user agent. Spoofed user agents inflate numbers.
- Look at errors per page, not just totals. A page that returns 403 to one bot and 200 to another has a WAF rule problem.
Promptwatch's AI crawler logs cover GPTBot, OAI-SearchBot, ChatGPT-User, ClaudeBot, PerplexityBot and others, with plain-language explanations of crawl failures. DarkVisitors is another option for bot and agent visibility.

Profound's Pages feature also puts bot activity next to citation data, with a Bot View that compares what a crawler sees against what a human sees. That is handy for catching content that looks complete in a browser but reads as empty to a bot.
Stage 5: cited
A crawl does not guarantee a citation. Engines fetch many pages and use few. Profound describes this as pages that show up as retrieval candidates but rarely earn a citation, which points to a content issue rather than a visibility one. I like that framing because it stops you from fixing the wrong thing.
What to record per cited page
- First cited date
- Which prompts and which models cite it
- Citation position in the source list
- Whether it is still cited a month later
Sources vary in how they handle this. LLM Pulse's help documentation notes that a cited URL stays in its table and counts toward citation metrics even when page content cannot be read, and that 404 and 410 statuses stay visible as page status. It also re-crawls general pages on a 28-day interval. That matters if you are comparing a cached copy against your live edit: the copy can be a month old.
Citations decay, so keep tracking
The first citation is not the end. Promptwatch Data's content half-life report took every classified page ChatGPT cited on July 6, 2026 and followed them for 90 days.
| Content type | Still cited after 30 days | Still cited after 60 days |
|---|---|---|
| Landing pages | 78% | 69% |
| Docs | 76% | 67% |
| Comparisons | 72% | 63% |
| Listicles | not reported | 60% |
| Product pages, reviews, how-tos | not reported | about 55% |
| News | not reported | 41% |
News is the outlier: almost a quarter of cited news pages are never cited again after day zero, and half are gone after 36 days. Product pages, reviews and how-tos have a half-life of 74 to 77 days. The report's advice is to treat landing pages, docs and pricing pages as long-term citation assets and keep prices and specs on them accurate, and to use news for short bursts.
For your tracking, this means a page's status should not flip to "done" when it is first cited. A cited product page needs a freshness check on a schedule.
Page-level citation tools
Several tools can give you URL-level citation data. Otterly stores every cited URL per prompt and engine and sorts changes into new, increased, decreased and lost. Semrush lists cited domains and URLs ranked by citation order. Ahrefs Brand Radar shows cited pages across a large prompt index.


Promptwatch's Page Tracker lets you watch any URL, yours or a competitor's, and see the prompts and responses that cite it. For a broader view of the category, the GEO software directory at bestgeosoftware.com lists more options.
Stage 6: clicked, and why it looks different
Here is where reporting goes wrong. People expect the pages that get cited to be the pages that get traffic. They are mostly not.

Aleyda Solis analyzed April 2026 US Semrush data for 40 top sites across travel, finance, real estate and retail. Her figures, which she says should be read directionally:
| Page type | Share of AI traffic | Share of AI citations |
|---|---|---|
| Homepage / brand entry | 57.7% | 3.0% |
| Search / browse / category / listing | 4.5% | 28.3% |
| Guide / informational / editorial | 0.2% | 12.3% |
| Support / help / policy / locator | 1.2% | 10.1% |
| Product / listing / property detail | 2.9% | 6.2% |
| Account / app / cart / booking / checkout | 5.3% | 1.3% |
| Auth / operational / redirect | 14.6% | 0.1% |
| Other deep / non-standard URLs | 13.4% | 38.6% |
AI traffic was also tiny overall: about 0.19% of visits across the 40 sites, against 20.45% for organic search.
What I take from this: a cited guide with zero clicks is not a failure. It may be doing the work of getting your brand into the answer. And a traffic spike on a login or OAuth URL is not a sign your content strategy is working. In her dataset some of that traffic looks like agents following workflow URLs.
So track click data per page type, and do not roll everything into one AI traffic number.
Measuring the click side
You need a lightweight script, a Google Tag Manager template or a CMS plugin that classifies referrers from ChatGPT, Perplexity, Gemini, Claude, Copilot and others. One wrinkle for 2026: ChatGPT stopped sending a referrer on October 8, 2026, so tagging links with utm_source=chatgpt.com and counting those visits is now worth setting up. Promptwatch's guide to attributing AI traffic walks through it. Similarweb and Bear AI are also in this space.
Putting it together: a per-URL record
The simplest useful model is one row per URL with these fields:
url
status: draft | staged | published | live
published_at
first_crawled_at, crawl_errors
bots_seen: [GPTBot, OAI-SearchBot, ...]
first_cited_at, prompts_citing, models_citing
citation_position
ai_visits_30d, ai_conversions_30d
last_updated_at
Then you can ask pointed questions:
- Published more than two weeks ago, zero crawls: access problem. Check robots.txt, CDN and WAF rules, sitemap, internal links.
- Crawled, zero citations after a month: content problem. Compare against pages AI already cites for the same prompts, and check for gaps in answer structure.
- Cited, no traffic: probably fine for informational pages. Not fine for pages meant to convert, so check position and whether the answer gives away everything.
- Cited, then dropped: refresh it. A diff-style refresh beats a rewrite.
- Traffic, no conversions: a landing page or journey problem, not an AI problem.
Promptwatch's Search Insights overlays Google Search Console and Bing data with AI crawls, citations and clicks, so you can find pages that rank in Google but never get cited by AI. That one query pays for itself fast.
A weekly routine that takes about an hour
- Pull the list of pages published or updated in the last 30 days.
- Check crawl status for each. Flag anything with zero AI bot hits after 14 days.
- Check citation status. Flag pages that were crawled but not cited.
- Review cited pages that lost citations since last week.
- Look at AI-referred visits by page type, not in total.
- Turn flags into actions: fix access, refresh content, add internal links, or write a new page for an uncovered prompt.
The Action Board in Promptwatch generates a prioritized list like this automatically, and its Content Agent can write and publish the follow-up pages, then the same tracking picks them up. That closed loop is the main reason I would pick a platform over stitching four tools together.
Choosing tools for each stage
| Need | Options |
|---|---|
| One platform from publish to traffic | Promptwatch |
| Page-level citations next to bot activity | Profound |
| Cheap URL-level citation reports | Otterly.AI, Geoptie |
| Citation data inside an SEO suite | Semrush, SE Ranking, Ahrefs Brand Radar |
| Bot and agent detection only | DarkVisitors |

For a longer list of rank and visibility trackers, see ai-rank-tools.com.
Mistakes I see repeatedly
- Treating a mention as a citation. A brand can be named in the text while the footnote links to a directory. LLM Pulse puts it plainly: high citation rate with low mention rate means AI links to you without talking about you.
- Reading one run of a prompt as truth. Sources shift between runs, days and countries. Look at multi-week trends.
- Skipping crawler logs because analytics looks fine. Analytics cannot see bots.
- Reporting a single AI traffic figure that mixes homepage visits, checkout sessions and OAuth redirects.
- Letting cited pages go stale because the job was marked done.
Summary
Following a page from draft to citation means giving every URL a status that survives publishing. Draft and staged are yours. Published, crawled, cited and clicked are earned, and each has its own evidence. Check them in that order, because a failure upstream makes everything downstream meaningless. Start with crawler logs this week: if the bots are not reading your new pages, no amount of content polish will matter.




