How Agencies Ship 500 Optimized Images a Month Without Losing a Week

Every agency has the same conversation. The image work is scoped, everyone agrees it matters, and then it quietly does not happen — because forty images across five clients is a day and a half that nobody has, and the deliverable that gets cut is always the one no client explicitly asked for.

This is a production problem, not a knowledge problem. Here is where the time actually goes and what can be removed, with costs as of July 2026.

The real cost, itemised

Per image, done properly:

Step	Time	Automatable?
Decide what the image should show	30–60s	Partly — templates by page type
Source or generate it	20–90s	Yes
Review and redo if wrong	0–120s	No — judgement
Rename to a descriptive filename	20–30s	Yes
Write alt text	40–60s	Mostly — draft then skim
Add description, creator, copyright	30–45s	Yes
Add coordinates where relevant	15–30s	Yes
Resize, convert, compress	15–30s	Yes
Upload and place	30–60s	Partly

Call it three minutes average. At 500 images a month that is 25 hours — most of a working week, spent almost entirely on steps that require no judgement whatsoever.

Look at the automatable column. Everything except "decide what it should show" and "review it" is mechanical. Those two are where a human adds value; the other seven are where the week disappears.

Standardise before you automate

Automating an inconsistent process produces consistent inconsistency. Write these down first.

A filename convention. One pattern, applied everywhere: [subject]-[action]-[location?]-[variant?], lowercase, hyphenated. The reasoning is in SEO image filenames; what matters here is that it is written down and identical across the team.

An alt-text pattern per image type. Not a fixed string — a template with genuine variables. {action} on a {subject} in {city} produces accurate varied output as long as the variables actually vary. See how to write alt text for what to put in them.

A per-client context block. Business name, cities served, coordinates per city, service vocabulary, brand style notes, copyright string. Assembled once per client, reused on every batch. This is the artifact that makes batching pay.

A definition of done. Filename correct, alt text written, metadata block populated, correct format and size, placed on the page. If any step is optional it will be the one skipped.

The batch cycle

Once the standards exist, the monthly cycle is four steps.

1. Plan by campaign, not by page. List every image the month needs across the client's pages in one pass. Batching is only economical if the setup cost is paid once.

2. Write prompts that vary the scene. This is where quality is won or lost. Twenty images from one composition with twenty captions look exactly like what they are. Vary subject, angle, setting and time of day so the set reads as a set rather than a template.

3. Generate with the metadata attached. The pivotal decision. If your generator returns bare files, you have merely relocated steps 4 through 7 to later in the same week. If the filename, alt text and EXIF block are written at generation time, those steps cease to exist.

4. QC by sample. Every image gets a thumbnail-scale glance for obvious defects. A random ten percent gets a proper metadata inspection — download the published file and confirm the fields actually survived your pipeline, because CDNs and CMS resizers routinely strip them.

Multi-location is where it compounds

The arithmetic gets ugly fast. Twenty images across twenty-five cities is 500 files, each needing its own city-bearing filename, its own alt text naming the right place, and its own coordinates.

Done manually that is not a week, it is a fortnight, and it is exactly the kind of repetitive work where a transposed coordinate puts your Plano set in Ohio and nobody notices for a year.

The only sane approach is a city list plus a template, applied programmatically across the batch. This is what multi-city batch generation is for: one run, correct per-market metadata, no twenty-five-times-repeated manual pass. In SEOpix that sits on the Pro and Agency tiers — the pricing page shows where multi-city geo-tagging and 100-image batches begin.

The handover problem

One more reason to write the standards down: staff turnover. Image conventions live in people's heads more than almost any other part of an SEO process, and when the person who "just knows" how files are named leaves, the convention degrades within two months.

A one-page standard plus a per-client context block survives that. It is also what makes the work delegable to a junior or a contractor without a supervision overhead that cancels the saving.

Honest cost comparison

For 500 images a month:

Approach	Direct cost	Internal time	Notes
Stock library + manual metadata	Subscription fee	~25 hrs	Generic imagery; competitors use the same shots
Freelance designer	Per-image or retainer	~5 hrs coordination	Best quality, slowest turnaround
Raw AI generator + manual metadata	API cost	~20 hrs	Cheap images, full manual tail
Generator with metadata built in	$99/mo Agency tier	~5 hrs	Judgement work only
Original photography	Shoot cost	Scheduling	Irreplaceable for real people and real work

Two caveats worth stating plainly. First, the time column is the dominant number for most agencies — 20 hours of skilled time costs considerably more than any tier in this table. Second, generated images do not replace real photography for anything that must be genuine. Your team, your premises, your completed jobs — those need real photographs, because faking first-hand experience undermines the exact signal you are trying to demonstrate. Generation is for illustration, concept and diagram work, which is most of what a content page needs and none of what an about page needs.

A 90-day rollout that actually sticks

Attempting all of this at once across every client is how the initiative dies in week three. Sequence it.

Days 1–30: one client, standards only. Pick a mid-sized client and write the three artifacts — naming convention, alt-text templates, context block. Do that client's month manually, following the standard exactly, and time yourself honestly. You now have a real per-image cost rather than an estimate, and a written standard tested against reality instead of drafted in the abstract.

Days 31–60: automate the mechanical steps for that one client. Whatever removes renaming, alt-text drafting and metadata writing from the human path. Run the same volume again and compare the timing against your month-one baseline. If the saving is not obvious at this scale, it will not appear at ten clients either — better to find that out on one account.

Days 61–90: roll out to the rest, one client per week. Each rollout is mostly assembling that client's context block, since the convention and tooling already exist. Resist the urge to do five at once; the failure mode is a half-migrated portfolio where nobody remembers which clients are on which process.

The reason to go slowly is that the standards are the valuable part and they only get tested by use. Tooling applied to an unwritten convention just produces inconsistency faster.

Reporting it so it survives the next budget review

Image work gets cut because it is invisible in reports. Three numbers fix that, and none of them require new tracking.

Coverage. What proportion of images on the client's key pages have a descriptive filename and real alt text? Start at whatever it is — often under 20% — and report the trend. It is the only metric here fully within your control.

Google Images performance. Search Console's Performance report has a Search type filter; set it to Image. Impressions and clicks from image search are a distinct line most agencies never show a client, and it is directly attributable to this work.

Accessibility posture. Missing-alt count from any automated audit, trending toward zero. Increasingly this is a compliance conversation rather than an SEO one, which makes it easier to defend at renewal than a ranking argument.

Set expectations honestly alongside those numbers: image indexing is slow, and improvements show over weeks. Reporting a coverage number that moved this month while image impressions have not yet is a more credible story than implying the two move together.

What good looks like after three months

The agencies that make this stick end up in roughly the same place. A written naming and alt-text standard that a new hire can follow on day one. A per-client context block maintained like any other client asset. One production block per client per month rather than a trickle. Sample-based QC, including one real check that metadata survived to the live file. And image work that appears in the report because it actually shipped.

The unglamorous truth is that none of that requires new SEO knowledge. Everyone already knows images should have descriptive filenames and real alt text. It requires removing the seven mechanical steps that made doing it correctly cost a week a month.

Start with the 12-item checklist to fix your standard, then decide which of those steps you are still doing by hand.

Ten images a month are free if you want to test the workflow on one client before committing — or get in touch about Agency volume and sub-accounts.