Content Strategy

All steps (reference)

1. Step 1. Set up the workspace

   Open Setup and create the workspace. Creation is gated on a finished site architecture crawl — if one has not run, Setup says so and links to it rather than failing silently.

   What to do: Project → Content Strategy → Setup → Create workspace

   Where: content-strategy.setup

   Expected (pass)

   • Setup explains the crawl prerequisite when it is unmet
   • Workspace creation succeeds once a crawl has finished

2. Step 2. Confirm who you actually compete with

   Setup offers candidates with the evidence behind each and you tick the ones that are right. Nothing is added without that tick. People reliably name the brand they measure themselves against rather than whoever is winning their buyers' questions, so a free-text box is the escape hatch here, not the front door.

   What to do: Project → Content Strategy → Setup → Competitors

   Where: content-strategy.setup

   

   Expected (pass)

   • Each candidate names its evidence, and the three sources stay distinguishable: Local Rankings and Search results are observed, business profile is labelled "inferred, not observed"
   • "Cited for your questions" lists domains the engines actually cited when answering YOUR questions — the only source derived from what you are measured on
   • A confirmed rival stops being offered once the list rebuilds
   • Nothing to confirm reads as "nothing to confirm right now", never an empty box

   Negative cases (must fail safely)

   • A rival with no website can still be tracked by name, and the list says keyword comparison cannot include them

3. Step 3. Read This Week

   This Week is the recurring visit: what changed, and what needs acting on. A quiet week is a legitimate outcome and is stated plainly.

   What to do: Project → Content Strategy

   Where: content-strategy.this-week

   

   Expected (pass)

   • Actions first, then movement, then detail
   • A quiet week reads as quiet rather than broken
   • Each question shows a verdict per engine — cited, not cited, or not measured — never a merged score across engines
   • "Not measured" appears when the engine named no sources; it is a gap, not a zero

   Negative cases (must fail safely)

   • A failed or skipped run leaves a gap for that period rather than recording a zero citation rate

4. Step 4. Work the rewrite checklist

   When an engine cites someone else, the pages it named are fetched and compared against yours. Where most of them carry something yours does not — a comparison table, an FAQ section, a named author, a materially longer treatment — that becomes one line of work, grouped by the page it applies to rather than by the question. Each line carries the count it came from and how much of the evidence could actually be read. Mark done, Snooze, or Not for us; finished work stays reachable behind Show completed, and items close themselves once the page carries what was asked for.

   What to do: Project → Content Strategy → What to change

   Where: content-strategy.this-week

   

   Expected (pass)

   • One page cited for several questions is ONE item that lists those questions, not one per question
   • Every item states its count ("4 of 5 cited pages have a comparison table; yours does not") and its coverage ("3 of 5 analysed directly, 2 characterised from the answer")
   • A page that has changed shows the date with plain counts either side — no percentage, and no verb joining them
   • Show completed (N) keeps finished work on the page; the number matches what the control reveals
   • A question that produced no checklist names its reason rather than leaving an empty box

   Negative cases (must fail safely)

   • A period whose checklist has not been built says exactly that — it must not claim you were cited
   • A viewer without manage permission sees the items but cannot mark them done
   • A page address that is not http(s) renders as text, never as a link

5. Step 5. Run a teardown when you have time

   Research is the monthly-cadence tab, deliberately kept out of the weekly view. A teardown reads each confirmed competitor's sitemap, lists what they publish, reads the titles across the whole inventory, and reads a capped subset in full. It then counts what they write about. The price is shown on the button before you press it — this is the most expensive single thing the feature does, and it is charged on the click.

   What to do: Project → Content Strategy → Research → Run a teardown

   Where: content-strategy.research

   

   Expected (pass)

   • The credit cost and the competitor count appear beside the button, before it is pressed
   • Each topic states BOTH numbers separately — how many rivals cover it, and across how many pages
   • A coverage line says how many pages were listed, how many had titles, and how many were read in full
   • A competitor that could not be read carries the reason on its row, never a bare zero
   • A partial run is shown, not hidden — some rivals are always unreadable and the rest is still usable

   Negative cases (must fail safely)

   • No confirmed competitors → the button is replaced by a link to Setup, not offered at a price for work that cannot run
   • A run already in flight → the page says so instead of offering a second
   • A viewer without manage permission → no button, and the action refuses if called directly

6. Step 6. Pick what to write next

   The teardown ends with two lists, and they are deliberately not one. Catch up is "your rivals cover this and you do not". Nobody is serving these is "there is demand and no vendor is answering it". They rest on different evidence — competitor pages on one side, search demand and unanswered questions on the other — so merging them would rank "four rivals beat you here" against "nobody serves this" as the same size of thing. Every line shows the counts it came from, and below them one short sentence of interpretation in a quieter style: it reads the counts and says what they mean for someone deciding what to write. It never supplies a count of its own, and it is not on the ranking path.

   What to do: Project → Content Strategy → Research → What to write next

   Where: content-strategy.research

   

   Expected (pass)

   • Two separate columns — catch-up and whitespace evidence are never mixed into one ranked list
   • Every row carries its counts ("2 of your competitors cover this, across 19 of their pages"), and no row carries a score
   • The summary line states the cap when one applied ("2 more met the bar but fell outside the top 2") rather than stopping silently
   • The AI sentence sits below the counts, visually quieter, and states no number of its own
   • A column that could not be measured says so in words — it never prints a zero

   Negative cases (must fail safely)

   • No Google connection or no prompt set → the whitespace column says it has not been measured, NOT "nobody is serving anything"
   • A teardown that predates this feature → the section is absent rather than showing empty lists
   • The model unavailable or switched off → the counts still render, just without the sentences