How it works
No magic. Observable signals, compared carefully.
OrganicIQ is built on data anyone could collect by hand — and would never have time to. Here is exactly what it does, in order, including the places where it stops short of a claim it cannot support.
We find the pages, the way search engines do.
When you add a competitor, OrganicIQ reads their robots.txt and follows the sitemaps it declares, including sitemap indexes and gzipped sitemaps. If a site has no usable sitemap, it falls back to a single homepage fetch and the internal links on it.
Discovery respects robots.txt rules for our user agent, keeps a politeness delay between requests to the same host, and stops at the page limit your plan allows. It never crawls without bound.
Only public pages. No logins, no paywalled content, no scraping of private areas.
- GET/robots.txtRead crawl rules and declared sitemaps
- GET/sitemap.xmlParse urlset or follow the sitemap index
- GET/sitemap-posts.xmlCollect URLs with their lastmod dates
- FILTERcontent URLs onlyDrop assets, feeds, paginated and account URLs
- STOREpage inventoryFirst seen, last seen, source and status
Each page is recorded, then compared with itself.
For every monitored page OrganicIQ stores the HTTP status, final URL after redirects, title, meta description, H1, canonical URL, word count and a content fingerprint. Snapshots are kept, so the history of a page is a record, not a single current value.
The next check compares the new snapshot with the previous one and writes a change event for anything that genuinely moved. Cosmetic differences — a changed dash, trailing whitespace — are normalised away before comparison so they do not become noise.
A page removal is only recorded when a fetch actually returns 404 or 410. Disappearing from a sitemap is treated as a reason to re-check, not as proof.
- 2h agoCompetitor ANew page
/guides/ai-search-optimisation
- 5h agoCompetitor BTitle changed
“SEO Platform” → “AI Search & SEO Platform”
- YesterdayCompetitor AContent expanded
+1,240 words (68% longer)
- 2d agoCompetitor CNew topic cluster
4 pages on entity optimisation
- 3d agoCompetitor DPage removed
/features/rank-tracking returns 410
Topics come from what is on the page.
OrganicIQ extracts candidate phrases from titles, H1s, headings, meta descriptions and URL structure, then clusters phrases that describe the same subject. Clustering is conservative: a phrase joins a cluster only when it is a subset of that cluster and appears on substantially the same pages.
Your site is processed the same way, so comparing coverage is a database query rather than a guess. Each territory shows your pages, theirs, how many competitors cover it and how recently they moved — judged against the average pages per competitor, not the total, because summing would make any popular subject look like a gap.
A subject no monitored competitor covers is marked uncontested, not strong. Saying “strong” would imply we examined a market, when what we examined is a list of competitors you chose.
Missing topics
18
Shared competitor topics
11
Emerging topics
4
What actually appears, and what that tells us.
For each query you track, OrganicIQ records which pages the search provider returns, which of them belong to competitors you monitor, and what changed since the previous reading — new entrants, and pages that stopped appearing.
Search intent is read from the pages themselves: the URL shape first, because it is the most stable signal and the hardest to rewrite, then the title. A reading where no intent reaches a majority returns no intent rather than a coin toss.
The first provider reports relevance order, not ranking positions — so OrganicIQ records search presence and says so, and never shows a position nobody measured.
We ask the question and record the answer.
You write the prompts your buyers actually use. OrganicIQ sends each one to the provider and model you choose, stores the full response, and analyses which of your tracked brands it names, in what order, which domains it refers to and which sources it cites.
Brand matching is strict on purpose: whole-word matches on the brand name, plus the apex domain. A false “you were mentioned” is worse than a miss, so there is no fuzzy matching. Citations are separated into those the provider attached to its answer and those found in the prose, because the first is a statement about what it consulted and the second is a link a model typed.
Share of visibility is a share of mentions within the prompts you track, and the product says so wherever it shows one. With no mentions at all it reports none, rather than a row of zeros.
Tracked prompts
24
Prompt coverage
29%
AI mention gaps
13
Provider and model are always shown with every result. OrganicIQ measures the AI providers configured on your installation — not the whole AI ecosystem.
Two of your pages competing is a problem too.
OrganicIQ compares your pages against each other, not only against the market. Three signals have to agree before anything is reported: shared subject, shared wording, and the same apparent purpose.
That last one does the real work. A guide and a pricing page on one subject serve different needs — they are a funnel, not a conflict, and advising a merge there would be actively bad advice. OrganicIQ also stores the internal links it reads on each page, so a link suggestion can be checked against what already exists instead of guessed at.
Nothing is ever changed on your site. OrganicIQ reports the overlap and the evidence; merging pages is your decision.
Evidence becomes a recommended action.
Each opportunity carries an action — create, update, expand, merge, differentiate or monitor — and the argument for that action rather than another. The decisive input is almost always whether you already have a page on the subject, because recommending a new page when a competing one exists is how a site ends up fighting itself.
An observation nobody can act on is marked monitor and kept apart from the work, so the count of outstanding work is honest. An AI mention gap is one of those: no provider can be made to name a brand, and the product says so instead of inventing a task.
Every recommendation shows its reason in plain words. A label on a dashboard without an argument is an instruction, and this product does not issue those.
Create
No page of yours covers the subject. There is nothing to rework.
Update
A page of yours covers it. Rework that, rather than adding a second.
Expand
Several of your pages cover it thinly. Add depth, not another entry point.
Merge
Two of your pages compete on it. Resolve that before adding a third.
Differentiate
Competitors are established. Take an angle rather than matching them.
Monitor
Worth knowing. Nothing to do about it yet.
A brief from your data, then a measurement.
A content brief is assembled from what OrganicIQ has collected — competitor pages, topic clusters, search readings, AI citation sources, your own pages and the link graph. No model is asked what the page should say, so a brief costs nothing to make and every line of it traces to something you can check. A section with no evidence says so rather than being filled with plausible text.
Mark the work done and OrganicIQ records what was true at that moment, then compares it again after 7, 30 and 90 days. The baseline is taken before the work, never backfilled, because a “before” figure read afterwards would be a fabrication.
A metric OrganicIQ cannot read is reported as unknown, never as zero. “No change” and “never measured” are different answers.
Recommended intent
From the pages that appear for your tracked queries.
Supporting topics
From the topic cluster, not from a model.
Questions to answer
From headings read on competitor pages.
Internal links
Pages of yours that should link here and do not.
Cannibalisation warning
Shown before the structure, because it changes the plan.
What it was read from
Every section traces to stored evidence.
Scheduling and providers
Scheduled work
- Competitor discovery, snapshots and change detection run on a schedule you control.
- Search readings run weekly or monthly, or when you ask. Each one is a paid request, so nothing runs by surprise.
- Topics, gaps, own-page overlap and link suggestions are recomputed from stored signals — no network requests at all.
- AI prompts run manually, weekly or daily. Citation intelligence is recomputed from the answers already collected.
- The action queue and the weekly brief are built last, from everything above.
Every job is budgeted — a time limit, a page limit and a politeness delay — so monitoring never slows the application down, and nothing heavy ever runs while you are loading a page. A run that reaches its limit stops cleanly and the next one resumes where it left off, so coverage stays fair across everything you monitor.
AI providers
OpenRouter
One key, many models. Tracked prompts are answered by the models you enable — Claude, GPT, Gemini, Llama and others — through the OpenRouter gateway.
Providers are modular. Adding another one is a single implementation file — the rest of the product does not change, and no provider is ever a dependency of the whole application.