Finding & Merging Duplicates

As you add records, the same person often turns up in several of them. Someone might appear in a birth record and again in a census taken twenty years later. Each record creates or updates a person profile, so the same individual can end up split across two or more profiles. KleioBase automatically looks for profiles that are likely the same person and suggests merging them.

KleioBase never merges automatically. It surfaces a suggestion and shows you the evidence, and you always review and decide. The profiles being matched come from Uploading Records and from the people already in Your Knowledge Base.

Before you start

Duplicate matching is available on every plan, including the free Explorer plan. It runs automatically in the background as you add records - there is nothing to turn on. See Plans & Billing for what each plan includes.

Matching today works only within your own knowledge base. Matching against other users' trees (community matching) is not available yet.

Where suggestions appear

When KleioBase finds a likely duplicate, it shows up in a few places so you can act on it wherever you are:

• On your dashboard, as a count of potential duplicates. Clicking one opens that person's profile with the match ready to review.
• On the People page.
• In a "Potential Duplicates" section at the top of any profile that has matches. The number beside the heading is the full count of open matches for that person, even when only the first ten are listed. Use the "Show N more" button under the list to bring in the next ten.
• In your notifications, when a new Strong or Possible match is found. Weak matches do not send a notification, but they still appear everywhere else on this list.

You can start from any of these. Open the match to see who the two profiles are and why KleioBase thinks they are the same person.

When one person has been split across many records, you can end up with a dozen or more copies that all match each other. Work through them from a single profile: each merge folds the absorbed profile's own matches into the one you kept, so the count drops as you go and you never have to revisit the same pair twice. The count reflects every merge and dismissal straight away, so you do not need to reload the page to see where you are.

Reading the evidence

Every match comes with a confidence tier and a list of signals. The tier is a quick summary of how strong the match is. The signals let you judge for yourself.

Select "See more" on a match to see the suggested person in detail without leaving the page: their full birth and death dates and places, their parents, spouses, children and siblings, and the records they appear in. Two people with the same name and a similar birth year are often told apart only by their family, so check whether the parents and spouses line up with the person you are looking at. Close the window to compare against the profile behind it. The scoring signals are under "Why this was suggested".

Each match lists supporting evidence, such as a matching name, a matching birth year, a shared parent, or the same record appearing on both profiles. It also lists any contradicting signals, such as different names, a large birth-year gap, or places that are far apart. Read both sides before you decide. A high tier with one nagging contradiction is still worth a careful look, and a low tier with strong supporting evidence may still be a real match.

When names are spelled differently

Historical records spell the same name many ways. Aschgrow and Ashgrove, Yosef and Joseph, Warszawa and Warsaw are each one name written twice, and a plain letter-by-letter comparison cannot tell those apart from two names that merely look alike, such as Maria and Matilda.

When the automatic comparison is confident either way, KleioBase uses it. When it is genuinely unsure about a name or a place, the AI is asked one question about that pair of spellings: are these the same name, or not? If it does not know, it says so and nothing changes.

You will see the answer under the name or place line when you open a match and expand "Why this was suggested". For example, under "Similar last names" you might read that Aschgrow and Ashgrove are historically the same name, and under a first-name line that Irene and Erna are not. These explanations tell you why a name counted for or against the match, so you can disagree with the reasoning rather than just the result.

What counts as strong evidence

The tier reflects the actual documentary evidence connecting two people, not just how similar their names look. Two people who share a name but nothing else - no matching birth year, no shared family, no shared record - are held to the Weak tier rather than shown as a mid-confidence Possible match. A shared name is a starting point, not proof, and this keeps Possible and Strong meaning what they say.

Place evidence works the same way. Two people born in the same town, or in nearby towns, get real credit toward a match. Two people who were merely born in the same country or region do not, because that is true of a huge number of unrelated people and tells you almost nothing on its own. A shared country can still nudge a match a little, but it is never enough by itself to raise the tier the way a shared town, a shared parent, or a matching record is.

Confidence tiers

The tier is guidance, not a verdict. You make the final call.

How to merge two profiles

When you have decided two profiles are the same person, merging combines them into one.

1. Open the match and click Merge. This opens a side-by-side comparison of the two profiles.
2. Choose which profile to keep. This becomes the surviving profile. One is preselected for you: if you opened the match from a person's own profile page, that person is shown in the left column and is the one kept, since they are the person you were working on - and the same goes for their parents, spouses and children further down, so a merge started from one person keeps that person's side of the family by default. Opened from anywhere else - the dashboard, the Research Companion - the older of the two profiles is preselected instead. Either way you can switch by clicking the other card, on the main pair and on every relative independently.
3. For every field the two profiles can carry - name, dates, places, occupation, religion, burial, christening, cause of death, notes, and more - pick which value to keep. Below that, repeatable facts and timeline events from both profiles are listed as their own rows, so you can keep or drop each one individually. Each row that only one profile carries is tagged with what the other profile holds under the same label. "No equivalent" means the other profile has no entry under that label at all, so dropping the row loses the entry entirely. "Differs" means the other profile records that label with different content, so read the other rows under that label before deciding. Rows both profiles record identically are untagged. Only the ones the field list above does not already cover appear here - if a profile's occupation is also recorded as a fact with the same wording, you decide it once in the field list rather than twice. When the two profiles disagree on such a field - one says tailor, the other merchant - both wordings are listed here instead, because the merge keeps both facts and the field above only picks the one shown at the top of the profile. This lets you assemble the best version of the person from both profiles.
4. Review the relatives. Relatives that look like duplicates are already paired for you; mark any that are not actually the same person, and pair up any that were missed. See Reviewing relatives during a merge below.
5. Confirm to complete the merge.

If the two profiles look identical and you are not sure which is which, each one's ID is shown just above its card in the comparison. Click it to open that profile in a new tab, where you can compare its linked records before you decide.

Under each person's name, the comparison also lists their close family: their parents ("Son of Moses & Rivka"), their spouses, and their children. This is often the quickest way to tell two same-name people apart. For example, a man who married twice may have had a son with the same name from each marriage, and the parents line shows which mother each son belongs to. Someone recorded in two families gets a separate line for each set of parents. Any relative who also has a card on the page is underlined; click the name to jump straight to that person.

After you confirm, you land on the surviving profile so you can review the result.

What a merge does

A merge keeps one profile and removes the duplicate. It moves the other profile's records and relationships onto the profile you kept, and it combines their notes. The result is a single, richer profile with everything attached to it.

Details fill in the gaps. Anything the duplicate recorded that the profile you kept was missing - a death date, a place of burial, an occupation, a religion, a cause of death, a photo - is carried across automatically. Where both profiles have a value for the same detail, the one you kept wins unless you choose otherwise. Nothing you had already recorded is overwritten.

Facts and events you added by hand are combined too. If you noted a fact or a timeline event on one profile and something different on the other, the merged profile ends up with both. Entries that say the same thing on both sides are not duplicated.

If both profiles were linked to the same record - common, since appearing together in one document is often what flagged them as duplicates - the two links become one. A link carries one role, so when the two had different roles in that document, the family role wins: a man listed once as the groom and once as a witness stays linked as the groom, so the merge never turns a relative into a mere associate. If neither role is a family one, the role on the profile you kept is used.

Because the combined profile now holds more information, it gets re-checked for duplicates. A merge can therefore surface new matches that were not obvious before. This is normal. Review each new suggestion the same way.

Reviewing relatives during a merge

When you merge two people, KleioBase also looks at their relatives and lists them all in one Relatives section, grouped by relationship: parents, spouses, children, siblings, and other relatives. Each group says what is about to happen to it, for example "1 will be combined, 2 will stay separate", so you can see the outcome without opening every card.

Every relative is in one of two states.

Same person. The two records are the same individual and will be combined into one. Parents, spouses, and children that clearly line up start out this way, so the obvious pairings are already made for you. These show the same side-by-side comparison used for the main merge, so you can choose which details to keep.

Stays separate. Nothing is combined. That person stays in your tree as their own person, still connected to the profile you keep. This is the default for anyone who did not line up, and it is a safe outcome - nobody is deleted or hidden.

Each relative sits in the column of the profile it belongs to - the same left and right columns used for the two people you are merging - so you can always tell whose relative it is at a glance. The choices for that relative appear in the opposite column.

Changing a pairing

If a suggested pairing is wrong, press Not the same person. The pairing is undone and both people drop back to "Stays separate" with their own pickers, so you can point either one at somebody else - or put the original pairing back, since the person it was paired with is offered as a suggestion marked "was matched".

To pair two relatives yourself, use the controls beside the one you want to match. Relatives of the same type on the other side appear as quick suggestions, and the search box finds anyone else in your tree. Both are limited to a compatible sex, so a son is never offered as a match for a daughter. If someone you expect to see is missing from the suggestions and the search results, check that their sex is recorded correctly on their profile - a person recorded as male will not be offered as a match for a person recorded as female.

Choosing Don't keep (move to trash) sends that relative to the trash when the merge runs. You can restore them later from the Trash tab.

Undoing a merge

A merge is reversible. Open the profile you kept and choose "Undo merge" - this restores the people exactly as they were before, including any relatives you merged in the same step and any you moved to trash. Details that were filled in from the duplicate are cleared again, so the profile you kept goes back to exactly the values it had before the merge. One undo reverts the whole merge in a single step, so you can experiment with combining duplicates without worrying about getting stuck.

If you move the profile you kept to Trash and leave it there for 30 days, it is permanently deleted together with the duplicates that were merged into it, and that merge can no longer be undone.

Old links still work. If you follow a bookmark, or a link in an earlier Research Companion conversation, that points at a profile you have since merged away, you land on the surviving profile with a note explaining why. You never get a "not found" page for someone who is still in your tree.

Undo only reaches back to your most recent merge of that profile. If you merge a person, then later merge the result into someone else, undoing only reverses the later merge - the earlier one stays in place. So if you plan to undo, do it before stacking another merge on top.

Dismissing a suggestion

Not every suggestion is correct. Two different people can share a name and a rough time period and still be distinct individuals. When a suggestion is wrong, Dismiss it. A dismissed suggestion does not come back, so you will not have to judge the same pair again.

Dismissing is the right choice whenever the contradicting signals outweigh the supporting ones, or when you have outside knowledge that the two are not the same person.

If you dismissed a suggestion you actually wanted, nothing is lost - both people are still there, just no longer paired as a suggestion. You can still merge them by asking the Research Companion to do it: open the Companion from the sidebar, or use the "Ask Kleio" button on a person's profile, and tell it which two people to merge. If the two profiles disagree on a detail - two different birth years, say - it asks which value you want to keep before doing anything. It then shows you exactly what it is about to do, listing each detail it will keep, and waits for your confirmation before changing anything.

Merging two people who were never suggested

Matching only suggests pairs it considers likely, so two profiles you know are the same person will not always be offered as a match. The same Companion route covers this: ask the Research Companion to merge the two people, review the confirmation card it shows, and confirm. This works for any two profiles, whether or not a suggestion ever appeared.

Relatives are not suggested as duplicates

If two people are already connected as blood relatives in your tree, KleioBase will not suggest them as the same person. Being related means they are different individuals, so a parent and child, two siblings, or cousins (out to second cousins) are never offered as a duplicate of each other.

This also works after the fact. If a duplicate suggestion is already showing for two people and you then link them as relatives, the suggestion is removed automatically on the next matching run - you do not need to dismiss it by hand. If you later remove that relationship (for example, you linked the wrong people by mistake), the pair becomes eligible to be suggested again.

Because matching runs on a schedule, a suggestion you expect to disappear after linking relatives clears on the next run rather than instantly. Give it a few minutes and refresh.

When suggestions appear

Matching runs on a schedule in the background - roughly every 10 to 15 minutes. Because of this, suggestions do not appear the moment you make a change. After you add or edit people, confirm records, or import a tree, give it a few minutes and then check back. This is normal and expected.

If you confirm a record and do not see a match right away, it simply means the next scheduled run has not happened yet. Refresh after a short wait and the suggestion will appear.

A large import can surface matches over a longer period as KleioBase works through all the new profiles, so expect suggestions to keep arriving for a while after a big upload.

Facts recorded several ways

Not every duplicate is two profiles for the same person. Sometimes it is a single profile that records the same fact - most often an occupation - under several different wordings, because different records described it differently or one is a translation of another. KleioBase looks for this on the Researcher plan and above, as a "Same Fact Recorded Several Ways" conflict on the Conflicts tab.

For example, a profile might carry six Occupation entries where four are just translations of the same job and two are genuinely different jobs the person held. KleioBase groups the ones it believes say the same thing and asks you to pick which wording to keep.

Resolving the conflict

Open the conflict and you will see the grouped entries. Each one has a radio button for "this is the wording to keep" and a Same fact tick box, ticked by default. Pick the wording you want to keep, then press Keep "...".

The wording you kept becomes the value shown on the profile, and everything else in the group collapses into it. The collapsed wordings are not deleted: they are kept as alternates, shown as "also recorded as ...", so the source record each one came from still has its evidence on file.

A wording that matches the one you kept apart from capitalisation or spacing is not listed as an alternate, since "Tailor - also recorded as tailor" is not a finding. A wording that differs by more than that, including an accent, is kept: "Sastre - also recorded as Sastré" tells you a record spelled it differently.

If one entry in the group is not the same fact - a genuinely different job the person also held - untick its Same fact box before pressing Keep. It stays on the profile as its own separate fact, and KleioBase records that it and the wording you kept are distinct on this person, so the two are never grouped again.

If none of the entries belong together, use Keep all N separate. This leaves every entry as its own fact and never flags that group again, even as your data changes.

Seeing, adding and removing alternates

Alternates appear under their fact on the profile page as "also recorded as ...", and in the Facts tab of the profile editor on an Also row.

You do not need a conflict to use them. Every fact in the editor has an Also row with an Add wording box: type a wording, press Enter, and it is attached to that fact. This is the place to record that a relative was known as something the records do not say, or to add a reading from a document KleioBase has not seen. A wording you add by hand carries no source record, because it is your own note rather than something read off a document.

To remove an alternate, use the × beside it and save. Removing one does not affect the source record it came from. A wording that duplicates the fact's own value, or one already listed, is refused rather than added twice.

How the grouping works

KleioBase does not just check for identical spelling. When it does not already know whether two wordings mean the same thing, it asks Google's Gemini AI a narrow question: given these two values and the kind of fact they are, do they mean the same underlying thing? The answer only ever considers those two values and the fact type - never a name or any other detail from your tree - and every proposed answer is manually reviewed before it becomes part of the shared reference we use to decide.

Researcher plan and above

Facts recorded several ways is available on the Researcher plan and above. On the free Explorer plan, the conflict appears locked with a prompt to upgrade rather than showing the underlying facts; see Plans & Billing for what each plan includes.

Tips for working through duplicates

• Start with Strong matches. They are the safest to merge and clear the list quickly.
• For Possible and Weak matches, read the contradicting signals first. They tell you what to verify.
• Use the field-by-field choices during a merge to keep the most complete and accurate value for each detail, not just whatever was there first.
• In the Relatives section, keep the pairings you are confident about and mark the rest as not the same person. You can always merge those later when they appear as their own suggestions.
• After a big merge, expect a few fresh suggestions as the richer profile gets re-checked.

For more on building out the profiles these suggestions draw from, see Uploading Records and Your Knowledge Base. To work with your people directly, head to the People page.