When you scan, you can select between two different output options: You can either have editable text and images, or you will have a hidden text layer "behind" the original scanned image. From your description, it sounds like you have the latter, which means that all you are deleting is words from that hidden text layer - the image in the front remains unchanged. You would have to start over and re-run OCR, but with a different output format option to be able to delete/edit recognized text.
Thank you v. much for your answer. We are using Acrobat Pro DC for the first time to edit this pdf (urgently). Can you please help us understand exactly how to do these steps in Acrobat Pro DC on the scanned-in pdf that we were given: "start over and re-run OCR, but with a different output format option to be able to delete/edit recognized text". Thank you v. much.
By the way, if we try to use "enhanced scans", we get an error message with "this page contains renderable text". This file contains bookmarks that we need to preserve.
You can either start over with the original scanned PDF file (assuming that you've saved an unedited and un-OCRed version). Load it into Adobe Acrobat DC Pro and select Tools>Enhance Scans>Recognize Text, and then click on "Settings":
Here you can now select the "Editable Text and Images" format as output.
If going back to the raw scanned document is not an option, you can save your document as individual TIFF images. This will convert any "renderable text" to an image as well. You do this by selecting File>Export To>Image>TIFF - select a reasonably high resolution (e.g. 600 dpi). This will create one TIFF image per page in your document. You can now import these images again into Acrobat using File>Create>Combine Files into a Single PDF - then select all the TIFF images you just saved and bring them into the correct order and click on the "Combine" button. You can now run OCR again as explained before, and all text on your pages regardless of whether it was from the original scan or added as text later will now be recognized and converted to the "Editable Text and Images" format.