What are the pros and cons of using labels for every point on a survey scale?

Labels provide a semantic weighting to different points on a scale. If you fail to label every point, you may bias results in the following ways: 1️⃣ People will disproportionately choose points with labels 2️⃣ People may not know how to interpret unlabeled points.

Including a neutral or default option is, in my opinion, a completely separate issue from deciding whether or not to include labels.

I would like to see references for these claims, e.g., that labels increase cognitive load (like in comparison to what, no labels?). Good labels should reduce cognitive load, which should be established during cognitive pretesting. The risk of invalid data (e.g., in which respondents misinterpret response options) outweighs the risk of bias (which can be quantified and accommodated with statistical adjustments).

I'd also add that I would never have zero labels as you need to confirm the directionality of answers. If I don't tell you whether a 1 is low/bad or good/ high (and so on), then how can you accurately interpret the data? You need to at least label the ends of your scale. Perhaps you opt to do so by putting this in the text of your question; I'd want to reaffirm by including it in the scale itself (personal preference).

This is an important point. I have seen 4-point scales with "Excellent, Good, Fair and Poor." The difference between Good and Fair, conceptually, is not very much. But on a scale - when numbers are input to represent the labels, that difference is treated the same way as the difference between Fair and Poor. Most research suggests that anchors are best, but not in the way described below. Anchors in my background are labeling each of the end points (as is discussed below).

Labels do not have to always be balanced and symmetrical. It's well known that respondents often aggregate at the "top" of the scale, so you may want to have more discrimination there, e.g., excellent, very good, good, fair, poor.

The key is consistency and clarity. If you don't have labels will your participants know if 5=good or 5=bad? For data integrity don't make your participants guess or jump through hoops to give the answers you're seeking.

I have found a well thought out and clearly labeled rating scale produces the fewest errors with people getting the wrong end of the scale. It also enables more impactful and easy to understand reporting. i.e. the meaning of the labels provides a more intuitive understanding of what a 4.6 out 6 average really means vs just seeing the number alone. In cases where you don't have an outside benchmark to compare that number to I consider it essential.