Skip to main content
NB! Dokumentasjonen er automatisk generert. Informasjonen kan være ufullstendig og inneholde feil, spesielt skjermbilder og videoer. Se Om hjelpesidene. Vi vil veldig gjerne ha innspill: Kontakt oss via «Fant du det du lette etter?» nederst, i chatten nede til høyre eller på support@placepoint.no – vi svarer så fort vi kan!

Examples of using Placepoint AI

The reports below were made with Placepoint AI by our users, from a single prompt each. They show the range: from a lookup on one property to analyses of entire streets and portfolios. If you want to start with something short, go straight to the example prompts at the bottom of the page.

Area analysis: a whole street in one prompt​

The prompt "Check the whole of John Colletts allé, where can you build new, make a report" produced a 15-page report: all 72 land properties in the street reviewed, with a map and an assessment for each address.

The report found that the street is in practice three different planning realities, and that four addresses in the middle of the street are not covered by the building ban that stops all the neighbours. That exception is not in any map database: Claude found it by reading the planning documents and the decision on the extended building ban. The requirements are in the provisions, not in the colours on the zoning map. The method section lists which planning documents were actually read, and separates a documented gap in coverage from missing analysis. The same plans are in the zoning map layers in Placepoint Fusion.

Front pageSummaryZoning map
Front page of the report on John Colletts alléThe summary: the street is three streets, with key figures per zoneZoning map extract with land use areas over the garden city

The full report can be downloaded: John Colletts allé - zoning analysis (PDF, 15 pages).

More reports and analyses​

  • A 25-page sustainability report for a well-known head office in Oslo: matrikkel, BREEAM certification, ownership structure, all 25 tenants with a company dossier, registered ownership history, maps and a travel time isochrone.
  • An 11-page due diligence report with legal, technical and financial screening, which clearly marks what is covered and what is not.
  • A climate risk report as a PDF magazine with physical risk, transition risk and biodiversity, built on the same sources as the risk map layers, and a credit report for a bank made the same way.
  • A buyer assessment where Claude, from a completely neutral prompt, concluded that the risk was high because the buying company was in an ongoing court case.
  • A mass scan of land registers for a whole facility: Claude found all the matrikkel numbers itself, retrieved the extracts together and summarised the encumbrances that could block development.
  • A lead list for a roofing contractor: 1,517 properties identified and filtered on year of construction and building type, linked to the owning company and combined with visual roof analysis of buildings from aerial photos. The same kind of selection that Filter gives you in Placepoint Fusion.

Plot subdivision: which plots are large enough​

The question "Which plots in Bergsalléen in Oslo are large enough to be subdivided?" gives the answer in two parts: the rules that govern subdivision right now (area requirement per dwelling unit, %-BYA and the temporary ban with exceptions), and a table of the properties in the street that meet the area requirement, with matrikkel number and plot area. The answer is clear about what must be checked in detail with Plan- og bygningsetaten (the city planning and building authority) before an actual subdivision application.

Answer to the plot subdivision question for Bergsalléen, with rules and a table of plots that are large enough

This answer is from the Agent in Placepoint Fusion, which uses the same engine as Placepoint AI in Claude. The questions below work in both places:

  • "Check the whole of John Colletts allé, where can you build new, make a report."
  • "Which plots in Bergsalléen in Oslo are large enough to be subdivided?"
  • "Has any planning work been announced near Bergsalléen?"

How to get the most out of Placepoint AI​

Placepoint AI itself is already optimised at our end: which tools Claude picks, how the lookups are narrowed, and how much data comes back in each answer. That is not something you set yourself. What you control is the context Claude works in and the way you ask. That decides both the quality of the answer and how much of your Claude quota is used.

Give Claude instructions once, not in every prompt​

The simplest and most effective step is to tell Claude who you are, what you work on and what the deliverable should look like. Then you do not have to repeat it in every conversation, and the answers hit the mark from the first attempt. This applies to all use of Claude, not just Placepoint AI.

  • Your own profile: click your initials at the bottom left, select Settings and fill in Instructions for Claude. The instructions apply to all your conversations. Enter your role, company, field, the language you want answers in and the form the deliverable should take. See Understanding Claude's personalization features.
  • Projects: gather work that belongs together in a project, for example one client, one acquisition or one portfolio, and put the instructions and standing documents there. All conversations in the project inherit them. See What are projects? and How can I create and manage projects?.
  • The whole organisation: on Team and Enterprise plans, the Owner and Primary Owner can set instructions that apply to everyone in the business, under Organization settings, Organization and access. They take precedence over the individual user's own instructions where the two say something different. See Set organization instructions.

Anthropic's own advice is in Prompt engineering best practices, and the technical reference they keep updated is Prompting best practices. Four pieces of advice from there carry most of the load: say what Claude should do instead of what it should not do, give reasons for the requirements so the model understands the purpose, let it say that something is missing, and keep the instruction short. Longer instructions do not give better answers.

Example instructions for a business​

A law firm that uses Placepoint AI for factual groundwork, not for advice, can enter something like this for the whole organisation. The tags are not decoration: Anthropic recommends them because they keep the different requirements apart, and because you can later change one block without touching the rest.

<role>
You provide factual groundwork for lawyers. We make the legal assessment
ourselves, so the answer should be verifiable rather than conclusive.
</role>

<sources>
State the register and the date for every factual claim about a property, and
separate information from Grunnboken (the Land Registry) from matrikkel
information. Mark your own reasoning with "Inference:", so we can see what the
register says and what you derive. If something is not in the register, write
that it is not registered.
</sources>

<when_something_is_missing>
If you are missing context that materially changes the answer, ask one precise
question before you answer. Reasonable inferences are fine when they are
marked as such. Numbers, names, quotes and source references must come from
data you have retrieved.
</when_something_is_missing>

<personal_data>
You do not show personal names or national identity numbers: Placepoint AI
removes them. If we need an identified title holder, we look it up in
Placepoint Fusion, where access requires a legitimate interest, or in
Grunnboken.
</personal_data>

<delivery>
Answer with plain text or a table that we paste into our own templates. Ask
first if the task looks like it needs a report, PDF or dashboard: that takes
longer, and we rarely use it. End with findings, uncertainties and what should
be investigated further.
</delivery>

Notice what the instruction does: it says what Claude should do rather than what it should not do, it gives reasons for the requirements, and it gives the model a way out when the basis is thin, instead of forcing an answer. Those are the three moves that make the biggest difference.

Replace the content with what applies to your industry. A property company will for example ask for portfolio figures per SPV, a bank for a traffic light per risk topic.

Standard instructions by industry​

The blocks below are the same four moves applied to each industry: the role, what to prioritise, and the form the deliverable should take. The examples are short, for illustration. Pick your industry, and update the example with instructions that fit your work and your preferences.

<role>
You provide factual groundwork for collateral and credit assessment. We make
the credit decision ourselves.
</role>

<prioritise>
Encumbrances, title holder and mortgages first, then natural hazards such as
flood, landslide and quick clay. State the register and the date for each
piece of information.
</prioritise>

<delivery>
A table with one row per property: encumbrance, risk, source, date. Write "not
covered by the register" where coverage is missing, rather than leaving the
row out.
</delivery>

If several of you work in different ways, put the shared part at organisation level and the industry block in each project.

Ask for what you actually need​

  • One task per conversation. Start a new conversation when you change topic. A long conversation carries along everything retrieved earlier, and that costs both time and quota without making the answer better.
  • Say what form you want. If you need three numbers, ask for three numbers. A report, table, map or dashboard costs more than a short answer. Reports are delivered as HTML by default because that is fastest; ask for PDF when you actually need to send the document on.
  • Narrow it down. One matrikkel number, one street, one kommune or one organisation number gives faster and more precise answers than "Oslo".
  • Reuse the recipe. When an analysis works well, ask Claude to write the prompt down as a recipe you can run on the next property or portfolio. Put it in the project instructions.
  • Pick the model to fit the task. Use the most advanced model to build the structure and the recipe, and a faster model to run that same recipe many times.

Ask for sources, and read the caveats​

  • Require a source and a date for every factual claim. Claude will readily state which register the information comes from and when it was retrieved, and link to the map layer or the planning document. Ask for it as standard, so the answer is verifiable.
  • Ask Claude to separate what the register says from what it infers. That is the most important distinction in factual groundwork.
  • Definitions vary between kommuner. "Undeveloped" and "sentrumsformål" (central area purpose) do not mean the same thing in every kommune. Ask Claude to flag which definition has been used when the answer covers more than one kommune.
  • If data is missing, it should say so. Placepoint AI tells you when there are no planning data for a property, instead of describing it as unzoned.
  • Claude can be wrong. The results are a starting point, not a definitive answer, and must be verified against official sources before they are used for decisions.

If the connector has been added by an administrator for the whole business, each user must still log in with their own Placepoint account the first time. Access follows the user, not the URL. The steps for connecting are on Placepoint AI.

Short example prompts​

Placepoint AI takes both short questions and complex tasks: one lookup on an address, or a whole street with a report at the end. The examples below are easy to write and still use a lot of Placepoint at once: the registers, the map layers and the analyses in the same answer. Feel free to start with "Use Placepoint MCP", write in the language you use yourself, and replace the square brackets with a real address, a matrikkel number or an organisation number:

  • "Retrieve the land register extract for [matrikkel number] and summarise which encumbrances are on the property."
  • "Check whether the property [matrikkel number] is in a flood or landslide exposed area, and show it on a map."
  • "Which zoning plans and municipal plans apply to the property [matrikkel number]?"
  • "Check the whole of [street]: where can you build new? Make a report."
  • "Which plots in [street] are large enough to be subdivided?"
  • "Has any planning work been announced near [address]?"
  • "Retrieve the provisions and the planning description for the plan that applies to [matrikkel number], and summarise what is allowed in terms of land use ratio, heights and purpose."
  • "Show all historical aerial photos around the property [matrikkel number], set up neatly with metadata."
  • "Find the best places to set up a new grocery store near [place]."
  • "Make a risk report for [address]."
  • "Make a risk assessment of organisation number [orgnr] as the buyer of the property [matrikkel number]."
  • "Make a portfolio report on all the properties [company] (organisation number [orgnr]) owns in [place/area]."
  • "How many active properties are there in [kommune]?"
  • "Which properties does [company] rent in [kommune]?"
  • "Show detached house density in [kommune] as a map."
  • "Export all properties with tenants in [kommune] to a spreadsheet."
  • "Make a ranked list of commercial properties in [kommune] with only one tenant, where the tenant has weak finances."
  • "Make a list of IT companies in Bergen with more than 20 employees that sit in office buildings."
  • "Show the largest tenants of [company] in Bjørvika."
  • "Make a tenant overview for [address] with how long they have been there."
  • "Which properties are within 300 metres of Jernbanetorget, and who owns them?"
  • "Find the 10 best premises for a distribution centre for a company that has to make deliveries physically in [area], ranked by driving time."
  • "Make a full portfolio overview for organisation number [orgnr]: all properties with geography and a map of the main cluster, ownership structure from the ultimate owner down to the SPVs with flags for deleted companies and companies with negative equity, all tenants per building with how long each tenant has been registered, and an ESG profile per building."

With Placepoint Dataset MCP, which works with the datasets in your projects:

  • "Make a dataset in the project [project] with all commercial properties over 5,000 m² in [kommune], with owner and year of construction."
  • "Upload our tenant list, link the rows to the matrikkel on address and show which ones got no match."
  • "Add kommune and grunnkrets (basic statistical unit) to each row in [dataset], and count rows per kommune."
  • "Which properties in [dataset] have a new title holder or a new mortgage in the past year?"
  • "Find the shortest route that visits all the points in [dataset] by car from [address]."
  • "Set the colour on [dataset] by year of construction, and hide the columns we do not use."

Several of the prompts have a fixed page in Placepoint Fusion. The tenants for one property are in Units and tenants, and for a whole portfolio in the tenant report. Selections on industry, employees and building type are what Filter does. Driving time and proximity are in Analysis.

Three of the prompts only work in Claude Desktop, Claude Code and Cowork: historical aerial photos, density maps and export to a spreadsheet. The Agent makes neither images nor files, so there the answer comes as text and map links. The reports and maps are built on the same sources as the rest of Placepoint Fusion: see Data and sources and Placepoint in numbers.

Sist oppdatert
NB! Dokumentasjonen er automatisk generert. Informasjonen kan være ufullstendig og inneholde feil, spesielt skjermbilder og videoer. Se Om hjelpesidene. Vi vil veldig gjerne ha innspill: Kontakt oss via «Fant du det du lette etter?» nederst, i chatten nede til høyre eller på support@placepoint.no – vi svarer så fort vi kan!