Microsoft Ads Keyword Research: Planner, Intent and Evidence

#01Which customer decision are you researching?

Begin with the task a buyer is trying to complete. A query about hiding cables on a standing desk may concern compatibility, safety, installation or appearance. An advertiser selling only adhesive clips should not assume that every cable-management query is a fit for its range.

Create an intent map with the problem, constraints, likely product or service, evidence needed and next action. For a merchant, distinguish category browsing, exact-product comparison, installation research and replacement-part demand. For bulk office orders, separate purchasing enquiries from job searches, general office advice and consumer single-item shopping.

Intent is a hypothesis, not a psychological diagnosis. A short query can have several meanings. Write competing interpretations and decide what a truthful ad and useful landing page would need to clarify. Do not call a person ready to buy merely because a keyword appears commercial.

The Google match-types and search-terms lesson covers transferable vocabulary. Here, the research output must reflect Microsoft's own estimates, actual destination targeting and available offers rather than a renamed Google keyword list.

#02How should you seed Microsoft Keyword Planner?

Microsoft documents Keyword Planner as a tool for discovering ideas and examining historical or forecast information. Seeds can include phrases, a website or categories. Use several seed types because each can reveal different language and omissions.

Start with customer descriptions, product attributes, support questions and authorised sales notes. Remove personal information before placing material in research tools. A product specification such as clamp installation can be useful; a customer's private email address or order history is not a keyword seed.

Website-based extraction has limitations. Microsoft's documentation notes that dynamically generated runtime JavaScript content is not read in the same way as directly accessible page content. Missing ideas may therefore reflect extraction limits rather than a complete absence of demand. Inspect what the tool actually understood before judging the business opportunity.

Save a research export before adding anything to an account. A tool's save-to-account function is an operational change, not simply a notebook action. In this course, the exercise output is an offline plan for review; publishing keywords or campaigns requires separate authorisation.

#03Which settings make an estimate interpretable?

Record the market, language, date range, device or other available targeting choices and seed method. A result without these settings is hard to compare. A seasonal product forecast for one country does not become a year-round estimate for another merely because the phrase is the same.

Do not treat a suggested bid as a fixed price that Microsoft will charge. Competition and volume indicators describe the tool's estimates within its context. They do not establish an acquisition cost, a conversion rate or a guaranteed number of impressions.

Export the raw evidence and create a separate decision sheet. This preserves the difference between what Microsoft returned and what the marketer concluded. If a reviewer challenges a priority, the original inputs remain available rather than having been overwritten by a subjective score.

#04How should you assess commercial fit?

Use a relevance review before a volume ranking. Ask whether the business sells the requested item, serves the market, can fulfil the implied deadline and has a page that answers the decisive question. A large estimate for an unsupported product is not a growth opportunity for the current campaign.

Evaluate ambiguity explicitly. An illustrative query such as office organiser could mean a physical desk product, a calendar application or a person arranging an event. Write the ad and destination that would make the intended interpretation clear. If ambiguity remains too costly or misleading, narrow the concept or hold it for further research.

Then assess economics. A product with low contribution may need a different campaign role from a higher-value bulk order. Do not assign a universal maximum CPC based on product price alone. Estimate an acquisition allowance from margin, then model how different conversion rates would affect the affordable click cost.

Label scenarios illustrative. For example, an illustrative EUR 18 allowable acquisition cost and a hypothetical 3% purchase rate imply EUR 0.54 per click before considering other complications. This is arithmetic sensitivity analysis, not a forecast that Microsoft will deliver that rate or price. Change the inputs and the result changes immediately.

#05How do observed search terms improve research?

After authorised campaigns have meaningful data, search-term reports can show how actual searches related to configured keywords or matching systems. Compare query, matched keyword, match behaviour, destination, cost and business outcome. A keyword is the advertiser's input; the query is observed demand within the report's coverage.

Microsoft's report is not a complete census of every search. Its documentation describes filtered reporting and warns that report totals can differ from broader impression totals. Avoid using unreported queries as evidence that an intent never occurred or that matching was perfectly controlled.

Classify available terms into useful demand, ambiguous demand, unsupported demand and policy or reputation risk. Add a reason and proposed action. Some terms deserve a better page or clearer ad rather than a negative keyword. Others reveal products the business genuinely does not offer and should be excluded after semantic review.

Keep research and enforcement separate. A term can be interesting for future product planning while being unsuitable for the current campaign. Record that distinction so an analyst does not repeatedly reintroduce excluded demand because it looks commercially attractive in a spreadsheet.

#06How do keywords, themes and AI matching differ?

Search keywords have configured match types, but current positive matching uses meaning rather than simple character matching. A research plan should describe intended query families and unacceptable interpretations, not just a list of exact strings.

Performance Max search themes are signals, not a conventional keyword targeting list. Do not paste the entire Search inventory into themes and then claim the campaign is restricted to those phrases. Their role and the campaign's broader automation need separate governance.

AI Max search-term matching adds another review layer in eligible Search configurations. It can extend beyond a manual keyword plan using the campaign's content and signals. Research remains useful for defining commercial boundaries, but keywords alone no longer describe every possible matched opportunity.

Use a translation column when planning across products: Search keyword and match type, Performance Max theme, landing-page requirement, negative consideration or AI Max control. One buyer need can inform several systems without those systems becoming operationally equivalent.

#07How should you prioritise the first research backlog?

Rank opportunities by commercial fit, evidence quality, page readiness, economic plausibility and controllability. Use a transparent rubric rather than an unexplained aggregate score. A high-fit term with modest evidence can be worth a limited test; a high-volume term with no suitable offer should remain unapproved.

Include the cost of learning. An opportunity requiring a new landing page, feed repair and a new conversion architecture is not comparable with one that can use an existing trustworthy product page. This does not make the first opportunity bad, but it changes the launch sequence.

Assign an owner and next evidence step to every priority. The step might be verifying product compatibility, reviewing a Planner estimate in the correct market or inspecting a current query cluster. “Research more” is not a complete action unless the missing decision and evidence are stated.

#08What should the research handover preserve?

Attach the raw export, selected targeting, rejected intents and the page-readiness decision to the handover. A colleague should be able to rerun the research in the same context and understand why the shortlist differs from a simple volume ranking. Keep estimates, observed queries and customer-language hypotheses labelled separately so later reporting does not upgrade assumptions into historical facts.

#09Worked example: Ledgerline

Ledgerline researches demand for a clamp-mounted cable tray. Its illustrative initial list mixes under-desk trays, adhesive clips, electrician services and cable-management software. Sorting by estimated volume would hide the fact that these searches require different offers.

The marketer creates an intent map and finds that standing-desk compatibility is commercially relevant but not yet answered on the product page. Instead of launching a vague ad, the product owner verifies the illustrative specifications and prepares a useful compatibility explanation. Service and software terms remain outside the current offer.

Walkthrough

1. Collect product and customer-language seeds without including personal records.
2. Run phrase, category and website research with the same documented market and period.
3. Preserve raw Planner output and mark missing site extraction as a limitation.
4. Group ideas by buyer decision, ambiguity and available destination.
5. Model acquisition-cost sensitivity with clearly illustrative inputs.
6. Create a Search test shortlist, a page-improvement backlog and a separate exclusion review.
7. Plan to compare reported query evidence with business outcomes after an authorised launch.

The result is not a prediction of winning keywords. It is a set of defensible hypotheses with explicit evidence gaps and a clear reason for each proposed test.

#10Common mistakes

Copying Google's highest-volume terms into Microsoft assumes identical destination demand and ignores source-market differences. Using Planner's suggested bid as a guaranteed CPC adds false financial certainty. Both errors can make a forecast look precise while its foundations remain weak.

Another mistake is equating low reported volume with no customer need. Extraction limitations, market choices, dates and report coverage all matter. Conversely, customer interviews can reveal useful language but do not prove a large paid-search opportunity. Different evidence sources answer different questions.

#11Try it yourself

Build a research sheet for one product family with at least four distinct buyer decisions. For each, include a seed, expected ambiguity, relevant page, market, commercial fit and missing evidence. If you have authorised Planner access, add an export; otherwise use clearly labelled hypothetical entries.

Select three test candidates and explain why each deserves investigation before a higher-volume rejected idea. Draft one page improvement and one negative-review candidate. Finish by writing what observed result would disprove your interpretation, including poor qualification, unsupported variants or a destination mismatch.

#12Quick quiz

1. Does a Keyword Planner suggested bid establish the price of future clicks?

Answer: No. It is contextual planning information, not a fixed price, guaranteed traffic level or acquisition-cost forecast.

2. Why can website-based research miss relevant ideas?

Answer: Microsoft's extraction has limitations, including runtime JavaScript content, so missing ideas do not necessarily mean that customer demand is absent.

3. Are Performance Max search themes ordinary Search keywords?

Answer: No. Themes guide an automated campaign and are not a conventional keyword restriction with equivalent match-type control.

#13What makes a keyword plan reviewable?

Preserve the research settings, raw evidence, buyer interpretation, offer fit and next test. A reviewer should be able to understand why a term was selected without trusting an unexplained score or a volume estimate alone.

Previous: Microsoft Ads Google Import: Mapping, Checks and Safe Cutover · Next: Microsoft Ads Match Types: Search Terms and Negative Controls

Or return to the Microsoft Ads for Marketers track.

Related: Microsoft Ads Strategy: Inventory, Intent and Economics · Microsoft Ads Match Types: Search Terms and Negative Controls · Microsoft Ads Structure: Locations, Language and Distribution

#14Common questions

Are Keyword Planner estimates guaranteed results? No. They are contextual planning information affected by the selected inputs, market and period, not promises of clicks, cost or profitable conversions.

Why might website research miss product ideas? Microsoft notes limitations in extracting runtime JavaScript content. Check the site input and use additional seeds before interpreting missing ideas as absent demand.

Can Search keywords be copied directly into Performance Max targeting? Not as an equivalent restriction. Performance Max search themes are signals, while Search keywords have their own matching and control model.