Can’t Identify That Mark? How AI Can Help with Pottery, Glass and Silver

A practical guide to using AI to identify pottery marks, glass signatures and silver hallmarks.

AI is only as good as the information you give it.

That is something I have learned through using AI to research pottery, glass and silver. Given clear photographs and the right supporting information, it can make connections within seconds. Given a blurred image or an incomplete mark, it can produce an equally confident—but completely wrong—answer. My experience of both is why I wrote this practical guide.

Most people do not begin researching a mark because they want a general lesson in ceramics, glass or silver. They have an object in front of them and a problem to solve. Who made it? Where did it come from? How old is it? Does the mark prove that it is what someone claims?

The front of an object usually gets all the attention. Turn it over, however, and the base may reveal a printed pottery backstamp, an impressed name, a faint glass signature or a row of tiny silver hallmarks. These can provide important clues, but the mark rarely tells the whole story.

When I first started researching decorative objects, finding an answer usually depended on access to the right specialist book. British pottery marks required one set of references, Scandinavian or Italian glass another, while silver hallmarks varied by country, assay office and period. These books were not always easy to find and could be expensive. Building a useful reference library took years.

AI has changed that process enormously. Unfamiliar lettering can be transcribed, characters translated and possible makers suggested before I have even begun searching museum collections, auction archives and specialist sources. For me, it has become an invaluable research tool.

This guide explains how to give AI the right information, check that it has read the mark correctly and decide whether its answer fits the object in front of you.

Start Here: Take Three Photographs

Before searching for a maker, give AI the evidence it needs. A blurred close-up of half a mark is rarely enough, however distinctive it may appear.

First, photograph the complete object from the front, back and side. Its shape, proportions, decoration and method of construction may support—or contradict—the identification suggested by the mark.

Next, photograph the entire base. The position of a mark and the surface surrounding it can be important. A tight crop removes the footrim, pontil, construction and wear that help place an object.

Finally, take a sharply focused close-up of every mark, number, signature and label. If several marks appear together, photograph the complete group before taking individual close-ups so that their original order and relationship are preserved.

Record the dimensions and weight. If the object has a lid, removable handle or other separate component, photograph and measure that as well. With a set, check whether all the pieces actually carry the same mark.

Use soft, even daylight for printed marks. Impressed pottery marks and engraved glass signatures often need light coming from the side. The shadows created by this low, raking light can reveal lettering that disappears under direct overhead light. On translucent glazed pottery, holding the base towards a strong light may reveal an impressed mark or date code that is almost invisible under normal lighting. Clear glass may need to be photographed against both a light and a dark background, while diffused light helps reduce reflections from silver.

This impressed date code on a Victorian Wedgwood majolica plate is almost invisible under normal light. Holding the base towards a strong light reveals the shallow characters clearly enough to photograph and check.

Avoid filters and excessive digital sharpening. Both can invent edges and alter faint lettering. AI may then identify the enhancement rather than the original mark.

Do not scrub a pottery base, polish silver or remove an old glass label before recording it. Dirt may need careful removal eventually, but a label, trace of gilding or area of wear can be more informative than a bright surface. Do not rub chalk or powder into a faint mark to make it visible; changing the light and taking several photographs is safer.

What to Check on Pottery, Glass and Silver

The basic photographs are the same, but each material provides different supporting evidence.

Pottery and Porcelain

Make sure the photograph of the base includes the entire footrim and any exposed clay. Its colour and texture, the way the foot has been finished and the relationship between the glaze and the mark can all help to confirm—or question—an attribution.

Record every number, painted initial and impressed symbol, even if it appears unimportant. A factory name, pattern number, shape number, decorator’s mark and date code may all appear on the same base, but they do not mean the same thing.

A green printed Coalport backstamp, dating broadly from about 1891 to 1920, and the separate hand-painted pattern number 74322. The two marks provide different information: the backstamp identifies and helps date the manufacturer, while the painted number records the pattern.

With unfamiliar characters, note the direction in which they appear and do not assume that a translation will identify the manufacturer. The mark might refer to an artist, kiln, studio or place. The clay body, glaze, decoration and method of manufacture must still agree with the proposed origin and date.

Glass

Turn the object slowly under a strong light before deciding that it is unmarked. Engraved and acid-etched signatures can be so faint that they are visible from only one angle.

An engraved signature on the base of a Flygsfors Coquille vase. Angling the clear glass towards the light makes the inscription easier to see and photograph.

Look beyond the signature. Photograph the rim, base, pontil, grinding, polishing, mould lines and applied elements. When no permanent mark is present, these construction details become the evidence AI must examine.

If a paper or foil label survives, photograph it before cleaning the glass. Record its colour, lettering, shape and position, and do not remove it. Even a damaged label may contain enough information to direct the research.

Silver and Silver Plate

Photograph the entire sequence of marks in its original order, followed by a close-up of each individual punch. Record exactly where each group appears. A lid, handle, blade or other detachable component may carry a different mark if it was made separately or replaced later.

The shape surrounding a set of initials can be as important as the initials themselves. The same is true of the typeface, case and outline of a date letter, so these details need to be sharp enough to examine individually.

Do not assume that every silver-coloured object is solid silver or that every stamp is a hallmark. Terms such as EPNS indicate silver plate. A single “925” may indicate a claimed silver standard, but on a British piece it is not, by itself, a complete UK hallmark. In the current British system, a complete hallmark has three compulsory components: a sponsor’s mark, a fineness mark and an assay-office mark. The sponsor identifies the person or company that submitted the article for hallmarking and is not necessarily the person who physically made it. This is explained by the Goldsmiths’ Company Assay Office and the UK Government’s hallmarking guidance.

A Barker Brothers silver-plated tea service. Its appearance alone cannot tell us whether it is solid silver or silver plate; the marks on the base provide the answer.

Ask AI in the Right Order

Once the photographs are ready, resist the temptation to begin with, “Is this made by…?” That wording encourages AI to look for confirmation of your suggestion.

Start by asking it to read only what is visible:

Describe exactly what you can see in this mark, including any letters, numbers or symbols that are uncertain. Do not identify the maker yet.

Then examine the original mark yourself with a magnifying glass or jeweller’s loupe. Check every letter, number and symbol against the AI transcription. A worn impressed pottery mark, faint glass signature or tiny silver punch can easily be misread from a photograph.

If you see something different, take another photograph from that angle and tell AI which part of its reading may be wrong. A single misread letter or symbol can lead to a completely different maker or date.

Once the transcription has been checked, ask AI to interpret the information according to the material.

For pottery, ask it to separate the possible factory or maker from pattern, shape, decorator and date marks:

Classify each mark on this base. Which part could identify the factory or maker, and which might be a pattern number, shape number, date code or decorator’s mark?

For signed glass, ask for alternative readings of the signature. With unmarked glass, begin with the construction rather than a famous name:

Describe the rim, base, pontil, mould lines, applied elements and apparent method of manufacture. Do not suggest a maker yet.

For silver, ask AI to consider the complete group of punches:

Describe each punch separately. Identify which may be a sponsor’s mark, fineness mark, assay-office mark, date letter or silver-plate mark. State clearly when a symbol is uncertain.

Only after these observations have been made should you ask:

What are the most plausible identifications, what evidence supports each one, and what details would rule them out?

This order separates observation from interpretation. It also makes uncertainty visible instead of allowing one doubtful letter to become the foundation of a confident story.

Check the Answer Somewhere Else

A convincing identification should lead to comparable marks on comparable objects. Check museum collections, official assay offices, factory archives, specialist publications and recognised collector organisations.

Auction archives can be extremely helpful, particularly when several properly catalogued examples agree, but an auction description should still be checked rather than treated as proof on its own. An unattributed sales listing is a lead, not evidence. Ten listings repeating the same description may all originate from one incorrect identification.

The practical route is straightforward:

Photograph the object → examine the material → transcribe the mark → check it with a loupe → consider possible identifications → verify the answer.

Where AI Most Often Goes Wrong

Poor image quality is the most obvious problem. A faint impression, reflection on silver or etched line in clear glass can easily be interpreted as a letter that is not there.

AI is also inclined to complete incomplete information. Once it decides that three blurred shapes form a particular name, it may build a detailed history around that reading. The history itself may be perfectly accurate while having nothing to do with the object being researched.

Visual similarity creates another trap. Crowns, stars, animals, flowers and monograms were used by numerous makers in different countries. Finding a roughly similar crown does not establish a pottery, glassworks or silversmith.

Online repetition can make the problem worse. Incorrect attributions pass between marketplace listings, dealer descriptions and image results. Once repeated often enough, they begin to look authoritative. AI can return this accumulated error fluently and with complete confidence.

There is also the problem of the leading question. If you ask whether an object is by a well-known maker, AI may concentrate on the similarities and overlook the differences. Asking what would disprove the identification is often more revealing than asking for further confirmation.

A confident answer is not necessarily a correct answer.

Why Handling the Object Still Matters

At the moment, this remains AI’s most important limitation.

A photograph cannot fully convey weight, surface or balance. It cannot feel the texture of an unglazed pottery foot, the finish of a glass base or the construction and wear of a piece of silver.

When you handle an object, you gather information almost without realising it. A ceramic body may be heavier or coarser than expected. The glass may not have the weight or finish associated with the suggested factory. Areas of wear on a silver object may reveal a different metal beneath the surface.

Some of this can be photographed, but not all of it. If the physical object and the suggested identification disagree, the answer must be reconsidered—however convincing the mark appears.

What If There Is No Mark?

The absence of a mark does not make research impossible, but it does require greater caution.

AI can compare the complete object with museum collections, catalogues and documented examples. Shape, construction, material and decoration may suggest a country, period, designer or factory. An original glass label may have vanished, a pottery may not have marked every piece, and very small or old silver objects may require specialist interpretation.

An identification based on appearance alone should be described accordingly. “In the manner of,” “comparable to” or “possibly” may be more accurate than presenting a maker as established fact.

Sometimes the honest answer is that the maker remains unidentified. That is not a failure of research.

Does Identification Establish Authenticity or Value?

No. A mark tells us what the mark claims. It does not automatically prove when, where or by whom the object was made.

Marks and signatures can be copied, added later or used on reproductions. Authenticity depends on the agreement between the mark, materials, construction, design, age and documented history of the object. If silver content materially affects value, an image-based AI opinion is not enough; the object may need to be examined or tested by a qualified professional.

Value is a separate question. It depends on condition, quality, rarity, provenance, size, desirability and the current market. AI can help locate comparable sales, but asking prices are not the same as achieved prices, and an apparently similar object may differ in an important way.

Identification is the beginning of valuation, not the valuation itself.

Do Specialist Books Still Matter?

Very much so. AI has made it unnecessary to own every book before beginning a search, but it has not made specialist knowledge redundant.

A good reference book does more than reproduce a mark, it has details not usually found on the internet. It may show how it changed over time, distinguish between factories or silversmiths with similar initials, explain date systems and place the object within its proper historical context. That depth is difficult to replace with a single online match.

AI is often most useful in helping me discover which specialist source I need. It can shorten the journey considerably, translate information and make obscure sources easier to find. The strongest identifications still come from comparing several forms of evidence.

← Προηγούμενη ανάρτηση