Build the question universe from observable sources
Start with the fixed questions already used for provider monitoring. Add exact queries from a current Google Search Console or Bing Webmaster Tools export, preserving impressions, clicks, CTR, position, page, country, device and measurement period. Then inspect the crawled website for question-form headings that reveal the language the publisher already uses to serve an audience.
Deterministic templates can suggest evidence-seeking questions from strong page titles, but those suggestions must remain labelled modelled discovery until measured demand or provider observations validate them. Provenance matters: a professional should be able to distinguish an exact customer-search phrase from a generated research prompt immediately.
Search impressions are search-platform exposure in the imported period. They are not AI prompt volume, total addressable demand, or a forecast.
Classify decision intent and journey stage
Group questions into practical intent categories such as commercial evaluation, comparison, category discovery, risk and trust, evidence seeking, implementation, and informational learning. Map each to awareness, consideration, decision, or activation. This reveals whether the portfolio over-measures definitions while missing pricing, procurement, risk, and comparison decisions.
Language support must extend beyond translated interface labels. Classification should recognise decision words in the analysed language, preserve Unicode query text, and use language-aware segmentation when matching questions to pages. A German pricing phrase, a Chinese comparison question, and an Arabic risk query should not be forced through an English-only token model.
Measure whether one page owns each question
For every candidate, compare the question with each crawled page’s title, description, headings, claims, and primary topic. Report the strongest page, relevance, page readiness, and closest competing internal page. Classify a content gap when no page clears the minimum relevance threshold, a weak owner when one page is only loosely aligned, and an ownership conflict when two pages are too close.
Ownership analysis changes the action. A content gap needs one canonical asset. A conflict needs consolidation or deliberate differentiation. A weak owner needs a bounded upgrade. A strong owner may need protection rather than another article competing for the same intent.
Preserve demand without double-counting it
Similar search phrases may match several monitored questions. Keep the fuzzy matches visible for diagnosis, but allocate each imported demand row to one best candidate when calculating portfolio and intent totals. Otherwise five related content ideas can each claim the same 5,000 impressions and inflate the business case before any work begins.
Show current impressions, clicks, CTR and position beside the prior equivalent period when available. A growing phrase with weak click capture and no credible owner can deserve attention even before provider evidence exists. It should still be labelled measured search demand plus a modelled prioritisation—not observed AI demand.
Join provider evidence only when the question matches
For a question already in the fixed portfolio, attach verified provider checks, citations, citation coverage, returned competitors, brand mentions and recommendation context. A newly discovered search query has no provider outcome until it is activated and tested. Do not borrow results from a vaguely related prompt or turn an unavailable check into zero visibility.
Public-site competitor benchmarks can add another evidence layer: target rank within the crawled comparison set, leading domain and page-relevance gap. Label this as public content competition rather than a private search or model ranking.
Use a disclosed opportunity formula
A useful prioritisation can combine measured demand, commercial intent, ownership gap, observed provider gap, public evidence competition and measured trend. Publish the weights and component scores. Confidence should be strongest when both measured demand and provider evidence exist, moderate when one exists, and directional when the candidate comes only from crawled language or a deterministic template.
The score is a decision aid, not a keyword-difficulty metric licensed from a private database. The useful output is the evidence beneath it: why the question exists, how much measured demand it received, which page owns it, what competitors appear, and how the team will verify the chosen action.
Promote questions without corrupting the baseline
Provider measurement has a practical capacity limit. Rank unmonitored candidates and promote only the most material additions. Adding a question versions the portfolio and creates a new comparison baseline. Older runs remain valuable history, but they must not be blended into like-for-like movement for a changed denominator.
- Review provenance. Prefer measured demand and published audience language over unsupported templates.
- Confirm commercial relevance. A large informational phrase may be less valuable than a smaller decision-stage question.
- Assign an owner. Publish or strengthen one canonical page before expecting reliable attribution.
- Activate deliberately. Add the exact question to the fixed portfolio and record the new version.
- Verify separately. Compare search performance in equivalent periods and provider citations on the unchanged question; neither proves the other.
What the professional report should contain
The final research register should expose the question, language, intent, journey, provenance, measured demand, unique allocated demand, trend, countries, devices, owner status, owner URL, relevance, page score, provider checks, citation coverage, brand narrative, public competitor evidence, opportunity formula, confidence, action and repeat test. A downloadable row-level export lets analysts challenge the ranking rather than trusting a black box.