A little SEO helper for generating missing image alt text

Missing image alt text is rarely the most intellectually demanding part of an SEO project. It can, however, become one of the more tedious parts rather quickly.

Three images? You are probably quicker doing them manually.

Nearly 1,500 images across several languages? That is where a small amount of sensible automation starts to earn its keep.

I have built a Codex skill with the exceptionally imaginative name Fill Missing Image Alt Text. It works with a Screaming Frog export, reviews the images and their page context, and produces suggested alt text together with a clean, import-ready file.

You can download the skill from GitHub, use it as it is, adapt it or improve it.

What the tool actually does

The starting point is Screaming Frog’s Images > Missing Alt Text > Inlinks bulk export. You can give Codex the resulting CSV or Excel file. A Google Sheet containing the export works as well.

The skill then:

• downloads and reviews each unique image;
• groups repeated uses of the same image;
• checks the source page and its surrounding context where necessary;
• recommends concise alt text in the source page’s language;
• identifies images that appear to be decorative;
• records its classification, confidence and supporting evidence;
• flags uncertain cases for human review; and
• creates a separate import-ready workbook.

The review workbook is deliberately more detailed than the eventual import file. It gives you enough information to challenge a recommendation rather than presenting an AI-generated answer as if it had descended from the mountain on a stone tablet.

The import-ready version removes the review columns and leaves you with the original structure and the proposed alt text.

Why not generate alt text directly inside Screaming Frog?

You can use an API integration in Screaming Frog to generate alt text. For a quick job, that may be entirely sufficient.

I built this workflow because I wanted more context and a more useful review process.

A filename can occasionally provide a clue, but it is weak evidence. A photograph called IMG_4027.jpg tells us very little. Even a descriptive filename does not tell us why the image appears on that particular page.

The same picture could be:

• an important photograph of a named speaker on an event page;
• a linked image leading to a speaker profile;
• a repeated thumbnail beside text that already says the same thing; or
• a decorative image that should have empty alt text.

Those cases should not automatically receive the same answer.

In one of my tests, the tool reviewed a photograph from an event in Nottingham where I had spoken. By using the image and page context, it identified Sophie Logan, me, the event, the location and the date. That is considerably more useful than “two people standing in front of a crowd”.

It is also why the workflow inspects each unique image visually and refers back to the source page when the identity, function, language or surrounding context is not clear.

The useful part is context, not volume

Generating 1,500 strings is easy. Generating 1,500 strings that are actually appropriate for their individual uses is the harder bit.

I tested the skill on the Waikay website, which had nearly 1,500 rows in the missing-alt-text export. The same images sometimes appeared on English, French and Spanish pages.

A blunt bulk process can easily assign one English or French description everywhere the image occurs. This workflow can create row-specific recommendations when the page language or the purpose of the image changes.

That does not mean it will always get the decision right. During testing, it reused a French recommendation on a Spanish page. I asked it to audit all repeated images across different language sections, and it found and corrected the wider pattern.

That mistake is worth mentioning because it illustrates the proper role of the tool. It removes repetitive inspection and drafting work. It does not remove the need for review.

Decorative images need a different decision

Not every missing alt attribute needs a written description.

If an image adds no information, has no function or merely repeats nearby text, an empty alt attribute may be the right answer. The import-ready workbook therefore leaves genuinely decorative recommendations blank. It does not insert the literal text alt="" into the spreadsheet.

This distinction matters for accessibility. The objective is not to fill every empty cell because a crawler has coloured it red. The objective is to give people useful equivalent information without forcing screen-reader users to listen to unnecessary descriptions.

When this workflow makes sense

For a handful of straightforward images, use your judgement and write the alt text manually. Starting an automated workflow would save no time and may create more review work than it removes.

The skill becomes more useful when:

• the export contains hundreds or thousands of rows;
• the same images are reused across multiple pages;
• the website has several language sections;
• image identity or purpose depends on page context;
• you need a documented review file before implementation; or
• you want an import-ready output without manually restructuring the results.

It is a helper, not an argument for automating a task simply because automation is available.

How to use it

The full installation and usage instructions are in the GitHub repository.

In short:

1. Export missing image alt-text inlinks from Screaming Frog.
2. Install the skill in Codex.
3. Supply the CSV, Excel file or Google Sheet.
4. Ask Codex to review the missing image alt text.
5. Review the recommendations and any flagged uncertainties.
6. Use the separate import-ready workbook for implementation.

You are welcome to download it, change it and give it your own voice. If you improve the shared version, you can submit the change through GitHub and I will review it.

Screaming Frog and Codex are becoming a rather useful combination for the less glamorous parts of technical SEO. If the crawl is exposing wider problems with indexation, rendering, architecture or implementation, a technical SEO audit is the more sensible next step. This little helper will save me time on larger projects. Hopefully, it saves you some as well.

If you prefer small tools that remove spreadsheet drudgery without pretending to replace judgement, you may also find my free keyword intent classifier useful.

---